API Documentation
datasets
This module contains the DataSample class, MultiModalArray, MultiModalSparseArray, MultiModalSparseInfo and MultiModalData, class The DataSample class encapsulates a sample ‘s components nbL and nbEx numbers, MultiModalArray class inherit from numpy ndarray and contains a 2d data ndarray with the shape (n_samples, n_view_i * n_features_i)
0 |
1 |
2 |
3 |
---|---|---|---|
xxxxxxxx |
xxxx |
xxxx |
xxxx |
xxxxxxxx |
xxxx |
xxxx |
xxxx |
xxxxxxxx |
xxxx |
xxxx |
xxxx |
xxxxxxxx |
xxxx |
xxxx |
xxxx |
xxxxxxxx |
xxxx |
xxxx |
xxxx |
xxxxxxxx |
xxxx |
xxxx |
xxxx |
xxxxxxxx |
xxxx |
xxxx |
xxxx |
xxxxxxxx |
xxxx |
xxxx |
xxxx |
xxxxxxxx |
xxxx |
xxxx |
xxxx |
MultiModalSparseArray inherit from scipy sparce matrix with the shape (n_samples, n_view_i * n_features_i)
- class multimodal.datasets.data_sample.DataSample(data=None, **kwargs)
A DataSample instance
- Example
>>> from multimodal.datasets.base import load_dict >>> from multimodal.tests.datasets.get_dataset_path import get_dataset_path >>> from multimodal.datasets.data_sample import DataSample >>> file = 'input_x_dic.pkl' >>> data = load_dict(get_dataset_path(file)) >>> print(data.__class__) <class 'dict'> >>> s = DataSample(data) >>> type(s.data) <class 'multimodal.datasets.data_sample.MultiModalArray'>
Input:
- Parameters
- datadict
- kwargsothers arguments
- Attributes
data
{ array like} MultiModalArrayMultiModalArray
- clear() None. Remove all items from D.
- copy() a shallow copy of D
- property data
MultiModalArray
- fromkeys(iterable, value=None, /)
Create a new dictionary with keys from iterable and values set to value.
- get(key, default=None, /)
Return the value for key if key is in the dictionary, else default.
- items() a set-like object providing a view on D's items
- keys() a set-like object providing a view on D's keys
- pop(key, default=<unrepresentable>, /)
If the key is not found, return the default if given; otherwise, raise a KeyError.
- popitem(/)
Remove and return a (key, value) pair as a 2-tuple.
Pairs are returned in LIFO (last-in, first-out) order. Raises KeyError if the dict is empty.
- setdefault(key, default=None, /)
Insert key with a value of default if key is not in the dictionary.
Return the value for key if key is in the dictionary, else default.
- update([E, ]**F) None. Update D from dict/iterable E and F.
If E is present and has a .keys() method, then does: for k in E: D[k] = E[k] If E is present and lacks a .keys() method, then does: for k, v in E: D[k] = v In either case, this is followed by: for k in F: D[k] = F[k]
- values() an object providing a view on D's values
- class multimodal.datasets.data_sample.MultiModalArray(data, views_ind=None)
MultiModalArray inherit from numpy ndarray
- Parameters
- datacan be
- dictionary of multiview array with shape = (n_samples, n_features) for multi-view
for each view.
- {0: array([[]],
1: array([[]], …}
- numpy array like with shape = (n_samples, n_features) for multi-view
for each view.
- [[[…]],
[[…]], …]
- {array like} with (n_samples, nviews * n_features) with ‘views_ind’ diferent to ‘None’
for Multi-view input samples.
- views_indarray-like (default= None ) if None
[0, n_features//2, n_features]) is constructed (2 views) Paramater specifying how to extract the data views from X:
views_ind is a 1-D array of sorted integers, the entries indicate the limits of the slices used to extract the views, where view
n
is given byX[:, views_ind[n]:views_ind[n+1]]
.
- Attributes
- views_indlist of views’ indice (may be None)
- n_viewsint number of views
- shapes_int: list of int numbers of feature for each views
- :Example:
- >>> from multimodal.datasets.base import load_dict
- >>> from multimodal.tests.datasets.get_dataset_path import get_dataset_path
- >>> from multimodal.datasets.data_sample import DataSample
- >>> file = ‘input_x_dic.pkl’
- >>> data = load_dict(get_dataset_path(file))
- >>> print(data.__class__)
- <class ‘dict’>
- >>> multiviews = MultiModalArray(data)
- >>> multiviews.shape
- (120, 240)
- >>> multiviews.shapes_int
- [120, 120]
- >>> multiviews.n_views
- 2
- T
The transposed array.
Same as
self.transpose()
.See also
Examples
>>> x = np.array([[1.,2.],[3.,4.]]) >>> x array([[ 1., 2.], [ 3., 4.]]) >>> x.T array([[ 1., 3.], [ 2., 4.]]) >>> x = np.array([1.,2.,3.,4.]) >>> x array([ 1., 2., 3., 4.]) >>> x.T array([ 1., 2., 3., 4.])
- all(axis=None, out=None, keepdims=False, *, where=True)
Returns True if all elements evaluate to True.
Refer to numpy.all for full documentation.
See also
numpy.all
equivalent function
- any(axis=None, out=None, keepdims=False, *, where=True)
Returns True if any of the elements of a evaluate to True.
Refer to numpy.any for full documentation.
See also
numpy.any
equivalent function
- argmax(axis=None, out=None)
Return indices of the maximum values along the given axis.
Refer to numpy.argmax for full documentation.
See also
numpy.argmax
equivalent function
- argmin(axis=None, out=None)
Return indices of the minimum values along the given axis.
Refer to numpy.argmin for detailed documentation.
See also
numpy.argmin
equivalent function
- argpartition(kth, axis=- 1, kind='introselect', order=None)
Returns the indices that would partition this array.
Refer to numpy.argpartition for full documentation.
New in version 1.8.0.
See also
numpy.argpartition
equivalent function
- argsort(axis=- 1, kind=None, order=None)
Returns the indices that would sort this array.
Refer to numpy.argsort for full documentation.
See also
numpy.argsort
equivalent function
- astype(dtype, order='K', casting='unsafe', subok=True, copy=True)
Copy of the array, cast to a specified type.
- Parameters
- dtypestr or dtype
Typecode or data-type to which the array is cast.
- order{‘C’, ‘F’, ‘A’, ‘K’}, optional
Controls the memory layout order of the result. ‘C’ means C order, ‘F’ means Fortran order, ‘A’ means ‘F’ order if all the arrays are Fortran contiguous, ‘C’ order otherwise, and ‘K’ means as close to the order the array elements appear in memory as possible. Default is ‘K’.
- casting{‘no’, ‘equiv’, ‘safe’, ‘same_kind’, ‘unsafe’}, optional
Controls what kind of data casting may occur. Defaults to ‘unsafe’ for backwards compatibility.
‘no’ means the data types should not be cast at all.
‘equiv’ means only byte-order changes are allowed.
‘safe’ means only casts which can preserve values are allowed.
‘same_kind’ means only safe casts or casts within a kind, like float64 to float32, are allowed.
‘unsafe’ means any data conversions may be done.
- subokbool, optional
If True, then sub-classes will be passed-through (default), otherwise the returned array will be forced to be a base-class array.
- copybool, optional
By default, astype always returns a newly allocated array. If this is set to false, and the dtype, order, and subok requirements are satisfied, the input array is returned instead of a copy.
- Returns
- arr_tndarray
Unless copy is False and the other conditions for returning the input array are satisfied (see description for copy input parameter), arr_t is a new array of the same shape as the input array, with dtype, order given by dtype, order.
- Raises
- ComplexWarning
When casting from complex to float or int. To avoid this, one should use
a.real.astype(t)
.
Notes
Changed in version 1.17.0: Casting between a simple data type and a structured one is possible only for “unsafe” casting. Casting to multiple fields is allowed, but casting from multiple fields is not.
Changed in version 1.9.0: Casting from numeric to string types in ‘safe’ casting mode requires that the string dtype length is long enough to store the max integer/float value converted.
Examples
>>> x = np.array([1, 2, 2.5]) >>> x array([1. , 2. , 2.5])
>>> x.astype(int) array([1, 2, 2])
- base
Base object if memory is from some other object.
Examples
The base of an array that owns its memory is None:
>>> x = np.array([1,2,3,4]) >>> x.base is None True
Slicing creates a view, whose memory is shared with x:
>>> y = x[2:] >>> y.base is x True
- byteswap(inplace=False)
Swap the bytes of the array elements
Toggle between low-endian and big-endian data representation by returning a byteswapped array, optionally swapped in-place. Arrays of byte-strings are not swapped. The real and imaginary parts of a complex number are swapped individually.
- Parameters
- inplacebool, optional
If
True
, swap bytes in-place, default isFalse
.
- Returns
- outndarray
The byteswapped array. If inplace is
True
, this is a view to self.
Examples
>>> A = np.array([1, 256, 8755], dtype=np.int16) >>> list(map(hex, A)) ['0x1', '0x100', '0x2233'] >>> A.byteswap(inplace=True) array([ 256, 1, 13090], dtype=int16) >>> list(map(hex, A)) ['0x100', '0x1', '0x3322']
Arrays of byte-strings are not swapped
>>> A = np.array([b'ceg', b'fac']) >>> A.byteswap() array([b'ceg', b'fac'], dtype='|S3')
A.newbyteorder().byteswap()
produces an array with the same valuesbut different representation in memory
>>> A = np.array([1, 2, 3]) >>> A.view(np.uint8) array([1, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0, 0], dtype=uint8) >>> A.newbyteorder().byteswap(inplace=True) array([1, 2, 3]) >>> A.view(np.uint8) array([0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 3], dtype=uint8)
- choose(choices, out=None, mode='raise')
Use an index array to construct a new array from a set of choices.
Refer to numpy.choose for full documentation.
See also
numpy.choose
equivalent function
- clip(min=None, max=None, out=None, **kwargs)
Return an array whose values are limited to
[min, max]
. One of max or min must be given.Refer to numpy.clip for full documentation.
See also
numpy.clip
equivalent function
- compress(condition, axis=None, out=None)
Return selected slices of this array along given axis.
Refer to numpy.compress for full documentation.
See also
numpy.compress
equivalent function
- conj()
Complex-conjugate all elements.
Refer to numpy.conjugate for full documentation.
See also
numpy.conjugate
equivalent function
- conjugate()
Return the complex conjugate, element-wise.
Refer to numpy.conjugate for full documentation.
See also
numpy.conjugate
equivalent function
- copy(order='C')
Return a copy of the array.
- Parameters
- order{‘C’, ‘F’, ‘A’, ‘K’}, optional
Controls the memory layout of the copy. ‘C’ means C-order, ‘F’ means F-order, ‘A’ means ‘F’ if a is Fortran contiguous, ‘C’ otherwise. ‘K’ means match the layout of a as closely as possible. (Note that this function and
numpy.copy()
are very similar but have different default values for their order= arguments, and this function always passes sub-classes through.)
See also
numpy.copy
Similar function with different default behavior
numpy.copyto
Notes
This function is the preferred method for creating an array copy. The function
numpy.copy()
is similar, but it defaults to using order ‘K’, and will not pass sub-classes through by default.Examples
>>> x = np.array([[1,2,3],[4,5,6]], order='F')
>>> y = x.copy()
>>> x.fill(0)
>>> x array([[0, 0, 0], [0, 0, 0]])
>>> y array([[1, 2, 3], [4, 5, 6]])
>>> y.flags['C_CONTIGUOUS'] True
- ctypes
An object to simplify the interaction of the array with the ctypes module.
This attribute creates an object that makes it easier to use arrays when calling shared libraries with the ctypes module. The returned object has, among others, data, shape, and strides attributes (see Notes below) which themselves return ctypes objects that can be used as arguments to a shared library.
- Parameters
- None
- Returns
- cPython object
Possessing attributes data, shape, strides, etc.
See also
numpy.ctypeslib
Notes
Below are the public attributes of this object which were documented in “Guide to NumPy” (we have omitted undocumented public attributes, as well as documented private attributes):
- _ctypes.data
A pointer to the memory area of the array as a Python integer. This memory area may contain data that is not aligned, or not in correct byte-order. The memory area may not even be writeable. The array flags and data-type of this array should be respected when passing this attribute to arbitrary C-code to avoid trouble that can include Python crashing. User Beware! The value of this attribute is exactly the same as
self._array_interface_['data'][0]
.Note that unlike
data_as
, a reference will not be kept to the array: code likectypes.c_void_p((a + b).ctypes.data)
will result in a pointer to a deallocated array, and should be spelt(a + b).ctypes.data_as(ctypes.c_void_p)
- _ctypes.shape
A ctypes array of length self.ndim where the basetype is the C-integer corresponding to
dtype('p')
on this platform. This base-type could be ctypes.c_int, ctypes.c_long, or ctypes.c_longlong depending on the platform. The c_intp type is defined accordingly in numpy.ctypeslib. The ctypes array contains the shape of the underlying array.- Type
(c_intp*self.ndim)
- _ctypes.strides
A ctypes array of length self.ndim where the basetype is the same as for the shape attribute. This ctypes array contains the strides information from the underlying array. This strides information is important for showing how many bytes must be jumped to get to the next element in the array.
- Type
(c_intp*self.ndim)
- _ctypes.data_as(obj)
Return the data pointer cast to a particular c-types object. For example, calling
self._as_parameter_
is equivalent toself.data_as(ctypes.c_void_p)
. Perhaps you want to use the data as a pointer to a ctypes array of floating-point data:self.data_as(ctypes.POINTER(ctypes.c_double))
.The returned pointer will keep a reference to the array.
- _ctypes.shape_as(obj)
Return the shape tuple as an array of some other c-types type. For example:
self.shape_as(ctypes.c_short)
.
- _ctypes.strides_as(obj)
Return the strides tuple as an array of some other c-types type. For example:
self.strides_as(ctypes.c_longlong)
.
If the ctypes module is not available, then the ctypes attribute of array objects still returns something useful, but ctypes objects are not returned and errors may be raised instead. In particular, the object will still have the
as_parameter
attribute which will return an integer equal to the data attribute.Examples
>>> import ctypes >>> x = np.array([[0, 1], [2, 3]], dtype=np.int32) >>> x array([[0, 1], [2, 3]], dtype=int32) >>> x.ctypes.data 31962608 # may vary >>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_uint32)) <__main__.LP_c_uint object at 0x7ff2fc1fc200> # may vary >>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_uint32)).contents c_uint(0) >>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_uint64)).contents c_ulong(4294967296) >>> x.ctypes.shape <numpy.core._internal.c_long_Array_2 object at 0x7ff2fc1fce60> # may vary >>> x.ctypes.strides <numpy.core._internal.c_long_Array_2 object at 0x7ff2fc1ff320> # may vary
- cumprod(axis=None, dtype=None, out=None)
Return the cumulative product of the elements along the given axis.
Refer to numpy.cumprod for full documentation.
See also
numpy.cumprod
equivalent function
- cumsum(axis=None, dtype=None, out=None)
Return the cumulative sum of the elements along the given axis.
Refer to numpy.cumsum for full documentation.
See also
numpy.cumsum
equivalent function
- data
Python buffer object pointing to the start of the array’s data.
- diagonal(offset=0, axis1=0, axis2=1)
Return specified diagonals. In NumPy 1.9 the returned array is a read-only view instead of a copy as in previous NumPy versions. In a future version the read-only restriction will be removed.
Refer to
numpy.diagonal()
for full documentation.See also
numpy.diagonal
equivalent function
- dot(b, out=None)
Dot product of two arrays.
Refer to numpy.dot for full documentation.
See also
numpy.dot
equivalent function
Examples
>>> a = np.eye(2) >>> b = np.ones((2, 2)) * 2 >>> a.dot(b) array([[2., 2.], [2., 2.]])
This array method can be conveniently chained:
>>> a.dot(b).dot(b) array([[8., 8.], [8., 8.]])
- dtype
Data-type of the array’s elements.
- Parameters
- None
- Returns
- dnumpy dtype object
See also
numpy.dtype
Examples
>>> x array([[0, 1], [2, 3]]) >>> x.dtype dtype('int32') >>> type(x.dtype) <type 'numpy.dtype'>
- dump(file)
Dump a pickle of the array to the specified file. The array can be read back with pickle.load or numpy.load.
- Parameters
- filestr or Path
A string naming the dump file.
Changed in version 1.17.0: pathlib.Path objects are now accepted.
- dumps()
Returns the pickle of the array as a string. pickle.loads or numpy.loads will convert the string back to an array.
- Parameters
- None
- fill(value)
Fill the array with a scalar value.
- Parameters
- valuescalar
All elements of a will be assigned this value.
Examples
>>> a = np.array([1, 2]) >>> a.fill(0) >>> a array([0, 0]) >>> a = np.empty(2) >>> a.fill(1) >>> a array([1., 1.])
- flags
Information about the memory layout of the array.
Notes
The flags object can be accessed dictionary-like (as in
a.flags['WRITEABLE']
), or by using lowercased attribute names (as ina.flags.writeable
). Short flag names are only supported in dictionary access.Only the WRITEBACKIFCOPY, UPDATEIFCOPY, WRITEABLE, and ALIGNED flags can be changed by the user, via direct assignment to the attribute or dictionary entry, or by calling ndarray.setflags.
The array flags cannot be set arbitrarily:
UPDATEIFCOPY can only be set
False
.WRITEBACKIFCOPY can only be set
False
.ALIGNED can only be set
True
if the data is truly aligned.WRITEABLE can only be set
True
if the array owns its own memory or the ultimate owner of the memory exposes a writeable buffer interface or is a string.
Arrays can be both C-style and Fortran-style contiguous simultaneously. This is clear for 1-dimensional arrays, but can also be true for higher dimensional arrays.
Even for contiguous arrays a stride for a given dimension
arr.strides[dim]
may be arbitrary ifarr.shape[dim] == 1
or the array has no elements. It does not generally hold thatself.strides[-1] == self.itemsize
for C-style contiguous arrays orself.strides[0] == self.itemsize
for Fortran-style contiguous arrays is true.- Attributes
- C_CONTIGUOUS (C)
The data is in a single, C-style contiguous segment.
- F_CONTIGUOUS (F)
The data is in a single, Fortran-style contiguous segment.
- OWNDATA (O)
The array owns the memory it uses or borrows it from another object.
- WRITEABLE (W)
The data area can be written to. Setting this to False locks the data, making it read-only. A view (slice, etc.) inherits WRITEABLE from its base array at creation time, but a view of a writeable array may be subsequently locked while the base array remains writeable. (The opposite is not true, in that a view of a locked array may not be made writeable. However, currently, locking a base object does not lock any views that already reference it, so under that circumstance it is possible to alter the contents of a locked array via a previously created writeable view onto it.) Attempting to change a non-writeable array raises a RuntimeError exception.
- ALIGNED (A)
The data and all elements are aligned appropriately for the hardware.
- WRITEBACKIFCOPY (X)
This array is a copy of some other array. The C-API function PyArray_ResolveWritebackIfCopy must be called before deallocating to the base array will be updated with the contents of this array.
- UPDATEIFCOPY (U)
(Deprecated, use WRITEBACKIFCOPY) This array is a copy of some other array. When this array is deallocated, the base array will be updated with the contents of this array.
- FNC
F_CONTIGUOUS and not C_CONTIGUOUS.
- FORC
F_CONTIGUOUS or C_CONTIGUOUS (one-segment test).
- BEHAVED (B)
ALIGNED and WRITEABLE.
- CARRAY (CA)
BEHAVED and C_CONTIGUOUS.
- FARRAY (FA)
BEHAVED and F_CONTIGUOUS and not C_CONTIGUOUS.
- flat
A 1-D iterator over the array.
This is a numpy.flatiter instance, which acts similarly to, but is not a subclass of, Python’s built-in iterator object.
See also
flatten
Return a copy of the array collapsed into one dimension.
flatiter
Examples
>>> x = np.arange(1, 7).reshape(2, 3) >>> x array([[1, 2, 3], [4, 5, 6]]) >>> x.flat[3] 4 >>> x.T array([[1, 4], [2, 5], [3, 6]]) >>> x.T.flat[3] 5 >>> type(x.flat) <class 'numpy.flatiter'>
An assignment example:
>>> x.flat = 3; x array([[3, 3, 3], [3, 3, 3]]) >>> x.flat[[1,4]] = 1; x array([[3, 1, 3], [3, 1, 3]])
- flatten(order='C')
Return a copy of the array collapsed into one dimension.
- Parameters
- order{‘C’, ‘F’, ‘A’, ‘K’}, optional
‘C’ means to flatten in row-major (C-style) order. ‘F’ means to flatten in column-major (Fortran- style) order. ‘A’ means to flatten in column-major order if a is Fortran contiguous in memory, row-major order otherwise. ‘K’ means to flatten a in the order the elements occur in memory. The default is ‘C’.
- Returns
- yndarray
A copy of the input array, flattened to one dimension.
Examples
>>> a = np.array([[1,2], [3,4]]) >>> a.flatten() array([1, 2, 3, 4]) >>> a.flatten('F') array([1, 3, 2, 4])
- getfield(dtype, offset=0)
Returns a field of the given array as a certain type.
A field is a view of the array data with a given data-type. The values in the view are determined by the given type and the offset into the current array in bytes. The offset needs to be such that the view dtype fits in the array dtype; for example an array of dtype complex128 has 16-byte elements. If taking a view with a 32-bit integer (4 bytes), the offset needs to be between 0 and 12 bytes.
- Parameters
- dtypestr or dtype
The data type of the view. The dtype size of the view can not be larger than that of the array itself.
- offsetint
Number of bytes to skip before beginning the element view.
Examples
>>> x = np.diag([1.+1.j]*2) >>> x[1, 1] = 2 + 4.j >>> x array([[1.+1.j, 0.+0.j], [0.+0.j, 2.+4.j]]) >>> x.getfield(np.float64) array([[1., 0.], [0., 2.]])
By choosing an offset of 8 bytes we can select the complex part of the array for our view:
>>> x.getfield(np.float64, offset=8) array([[1., 0.], [0., 4.]])
- imag
The imaginary part of the array.
Examples
>>> x = np.sqrt([1+0j, 0+1j]) >>> x.imag array([ 0. , 0.70710678]) >>> x.imag.dtype dtype('float64')
- item(*args)
Copy an element of an array to a standard Python scalar and return it.
- Parameters
- *argsArguments (variable number and type)
none: in this case, the method only works for arrays with one element (a.size == 1), which element is copied into a standard Python scalar object and returned.
int_type: this argument is interpreted as a flat index into the array, specifying which element to copy and return.
tuple of int_types: functions as does a single int_type argument, except that the argument is interpreted as an nd-index into the array.
- Returns
- zStandard Python scalar object
A copy of the specified element of the array as a suitable Python scalar
Notes
When the data type of a is longdouble or clongdouble, item() returns a scalar array object because there is no available Python scalar that would not lose information. Void arrays return a buffer object for item(), unless fields are defined, in which case a tuple is returned.
item is very similar to a[args], except, instead of an array scalar, a standard Python scalar is returned. This can be useful for speeding up access to elements of the array and doing arithmetic on elements of the array using Python’s optimized math.
Examples
>>> np.random.seed(123) >>> x = np.random.randint(9, size=(3, 3)) >>> x array([[2, 2, 6], [1, 3, 6], [1, 0, 1]]) >>> x.item(3) 1 >>> x.item(7) 0 >>> x.item((0, 1)) 2 >>> x.item((2, 2)) 1
- itemset(*args)
Insert scalar into an array (scalar is cast to array’s dtype, if possible)
There must be at least 1 argument, and define the last argument as item. Then,
a.itemset(*args)
is equivalent to but faster thana[args] = item
. The item should be a scalar value and args must select a single item in the array a.- Parameters
- *argsArguments
If one argument: a scalar, only used in case a is of size 1. If two arguments: the last argument is the value to be set and must be a scalar, the first argument specifies a single array element location. It is either an int or a tuple.
Notes
Compared to indexing syntax, itemset provides some speed increase for placing a scalar into a particular location in an ndarray, if you must do this. However, generally this is discouraged: among other problems, it complicates the appearance of the code. Also, when using itemset (and item) inside a loop, be sure to assign the methods to a local variable to avoid the attribute look-up at each loop iteration.
Examples
>>> np.random.seed(123) >>> x = np.random.randint(9, size=(3, 3)) >>> x array([[2, 2, 6], [1, 3, 6], [1, 0, 1]]) >>> x.itemset(4, 0) >>> x.itemset((2, 2), 9) >>> x array([[2, 2, 6], [1, 0, 6], [1, 0, 9]])
- itemsize
Length of one array element in bytes.
Examples
>>> x = np.array([1,2,3], dtype=np.float64) >>> x.itemsize 8 >>> x = np.array([1,2,3], dtype=np.complex128) >>> x.itemsize 16
- max(axis=None, out=None, keepdims=False, initial=<no value>, where=True)
Return the maximum along a given axis.
Refer to numpy.amax for full documentation.
See also
numpy.amax
equivalent function
- mean(axis=None, dtype=None, out=None, keepdims=False, *, where=True)
Returns the average of the array elements along given axis.
Refer to numpy.mean for full documentation.
See also
numpy.mean
equivalent function
- min(axis=None, out=None, keepdims=False, initial=<no value>, where=True)
Return the minimum along a given axis.
Refer to numpy.amin for full documentation.
See also
numpy.amin
equivalent function
- nbytes
Total bytes consumed by the elements of the array.
Notes
Does not include memory consumed by non-element attributes of the array object.
Examples
>>> x = np.zeros((3,5,2), dtype=np.complex128) >>> x.nbytes 480 >>> np.prod(x.shape) * x.itemsize 480
- ndim
Number of array dimensions.
Examples
>>> x = np.array([1, 2, 3]) >>> x.ndim 1 >>> y = np.zeros((2, 3, 4)) >>> y.ndim 3
- newbyteorder(new_order='S', /)
Return the array with the same data viewed with a different byte order.
Equivalent to:
arr.view(arr.dtype.newbytorder(new_order))
Changes are also made in all fields and sub-arrays of the array data type.
- Parameters
- new_orderstring, optional
Byte order to force; a value from the byte order specifications below. new_order codes can be any of:
‘S’ - swap dtype from current to opposite endian
{‘<’, ‘little’} - little endian
{‘>’, ‘big’} - big endian
‘=’ - native order, equivalent to sys.byteorder
{‘|’, ‘I’} - ignore (no change to byte order)
The default value (‘S’) results in swapping the current byte order.
- Returns
- new_arrarray
New array object with the dtype reflecting given change to the byte order.
- nonzero()
Return the indices of the elements that are non-zero.
Refer to numpy.nonzero for full documentation.
See also
numpy.nonzero
equivalent function
- partition(kth, axis=- 1, kind='introselect', order=None)
Rearranges the elements in the array in such a way that the value of the element in kth position is in the position it would be in a sorted array. All elements smaller than the kth element are moved before this element and all equal or greater are moved behind it. The ordering of the elements in the two partitions is undefined.
New in version 1.8.0.
- Parameters
- kthint or sequence of ints
Element index to partition by. The kth element value will be in its final sorted position and all smaller elements will be moved before it and all equal or greater elements behind it. The order of all elements in the partitions is undefined. If provided with a sequence of kth it will partition all elements indexed by kth of them into their sorted position at once.
- axisint, optional
Axis along which to sort. Default is -1, which means sort along the last axis.
- kind{‘introselect’}, optional
Selection algorithm. Default is ‘introselect’.
- orderstr or list of str, optional
When a is an array with fields defined, this argument specifies which fields to compare first, second, etc. A single field can be specified as a string, and not all fields need to be specified, but unspecified fields will still be used, in the order in which they come up in the dtype, to break ties.
See also
numpy.partition
Return a parititioned copy of an array.
argpartition
Indirect partition.
sort
Full sort.
Notes
See
np.partition
for notes on the different algorithms.Examples
>>> a = np.array([3, 4, 2, 1]) >>> a.partition(3) >>> a array([2, 1, 3, 4])
>>> a.partition((1, 3)) >>> a array([1, 2, 3, 4])
- prod(axis=None, dtype=None, out=None, keepdims=False, initial=1, where=True)
Return the product of the array elements over the given axis
Refer to numpy.prod for full documentation.
See also
numpy.prod
equivalent function
- ptp(axis=None, out=None, keepdims=False)
Peak to peak (maximum - minimum) value along a given axis.
Refer to numpy.ptp for full documentation.
See also
numpy.ptp
equivalent function
- put(indices, values, mode='raise')
Set
a.flat[n] = values[n]
for all n in indices.Refer to numpy.put for full documentation.
See also
numpy.put
equivalent function
- ravel([order])
Return a flattened array.
Refer to numpy.ravel for full documentation.
See also
numpy.ravel
equivalent function
ndarray.flat
a flat iterator on the array.
- real
The real part of the array.
See also
numpy.real
equivalent function
Examples
>>> x = np.sqrt([1+0j, 0+1j]) >>> x.real array([ 1. , 0.70710678]) >>> x.real.dtype dtype('float64')
- repeat(repeats, axis=None)
Repeat elements of an array.
Refer to numpy.repeat for full documentation.
See also
numpy.repeat
equivalent function
- reshape(shape, order='C')
Returns an array containing the same data with a new shape.
Refer to numpy.reshape for full documentation.
See also
numpy.reshape
equivalent function
Notes
Unlike the free function numpy.reshape, this method on ndarray allows the elements of the shape parameter to be passed in as separate arguments. For example,
a.reshape(10, 11)
is equivalent toa.reshape((10, 11))
.
- resize(new_shape, refcheck=True)
Change shape and size of array in-place.
- Parameters
- new_shapetuple of ints, or n ints
Shape of resized array.
- refcheckbool, optional
If False, reference count will not be checked. Default is True.
- Returns
- None
- Raises
- ValueError
If a does not own its own data or references or views to it exist, and the data memory must be changed. PyPy only: will always raise if the data memory must be changed, since there is no reliable way to determine if references or views to it exist.
- SystemError
If the order keyword argument is specified. This behaviour is a bug in NumPy.
See also
resize
Return a new array with the specified shape.
Notes
This reallocates space for the data area if necessary.
Only contiguous arrays (data elements consecutive in memory) can be resized.
The purpose of the reference count check is to make sure you do not use this array as a buffer for another Python object and then reallocate the memory. However, reference counts can increase in other ways so if you are sure that you have not shared the memory for this array with another Python object, then you may safely set refcheck to False.
Examples
Shrinking an array: array is flattened (in the order that the data are stored in memory), resized, and reshaped:
>>> a = np.array([[0, 1], [2, 3]], order='C') >>> a.resize((2, 1)) >>> a array([[0], [1]])
>>> a = np.array([[0, 1], [2, 3]], order='F') >>> a.resize((2, 1)) >>> a array([[0], [2]])
Enlarging an array: as above, but missing entries are filled with zeros:
>>> b = np.array([[0, 1], [2, 3]]) >>> b.resize(2, 3) # new_shape parameter doesn't have to be a tuple >>> b array([[0, 1, 2], [3, 0, 0]])
Referencing an array prevents resizing…
>>> c = a >>> a.resize((1, 1)) Traceback (most recent call last): ... ValueError: cannot resize an array that references or is referenced ...
Unless refcheck is False:
>>> a.resize((1, 1), refcheck=False) >>> a array([[0]]) >>> c array([[0]])
- round(decimals=0, out=None)
Return a with each element rounded to the given number of decimals.
Refer to numpy.around for full documentation.
See also
numpy.around
equivalent function
- searchsorted(v, side='left', sorter=None)
Find indices where elements of v should be inserted in a to maintain order.
For full documentation, see numpy.searchsorted
See also
numpy.searchsorted
equivalent function
- setfield(val, dtype, offset=0)
Put a value into a specified place in a field defined by a data-type.
Place val into a’s field defined by dtype and beginning offset bytes into the field.
- Parameters
- valobject
Value to be placed in field.
- dtypedtype object
Data-type of the field in which to place val.
- offsetint, optional
The number of bytes into the field at which to place val.
- Returns
- None
See also
Examples
>>> x = np.eye(3) >>> x.getfield(np.float64) array([[1., 0., 0.], [0., 1., 0.], [0., 0., 1.]]) >>> x.setfield(3, np.int32) >>> x.getfield(np.int32) array([[3, 3, 3], [3, 3, 3], [3, 3, 3]], dtype=int32) >>> x array([[1.0e+000, 1.5e-323, 1.5e-323], [1.5e-323, 1.0e+000, 1.5e-323], [1.5e-323, 1.5e-323, 1.0e+000]]) >>> x.setfield(np.eye(3), np.int32) >>> x array([[1., 0., 0.], [0., 1., 0.], [0., 0., 1.]])
- setflags(write=None, align=None, uic=None)
Set array flags WRITEABLE, ALIGNED, (WRITEBACKIFCOPY and UPDATEIFCOPY), respectively.
These Boolean-valued flags affect how numpy interprets the memory area used by a (see Notes below). The ALIGNED flag can only be set to True if the data is actually aligned according to the type. The WRITEBACKIFCOPY and (deprecated) UPDATEIFCOPY flags can never be set to True. The flag WRITEABLE can only be set to True if the array owns its own memory, or the ultimate owner of the memory exposes a writeable buffer interface, or is a string. (The exception for string is made so that unpickling can be done without copying memory.)
- Parameters
- writebool, optional
Describes whether or not a can be written to.
- alignbool, optional
Describes whether or not a is aligned properly for its type.
- uicbool, optional
Describes whether or not a is a copy of another “base” array.
Notes
Array flags provide information about how the memory area used for the array is to be interpreted. There are 7 Boolean flags in use, only four of which can be changed by the user: WRITEBACKIFCOPY, UPDATEIFCOPY, WRITEABLE, and ALIGNED.
WRITEABLE (W) the data area can be written to;
ALIGNED (A) the data and strides are aligned appropriately for the hardware (as determined by the compiler);
UPDATEIFCOPY (U) (deprecated), replaced by WRITEBACKIFCOPY;
WRITEBACKIFCOPY (X) this array is a copy of some other array (referenced by .base). When the C-API function PyArray_ResolveWritebackIfCopy is called, the base array will be updated with the contents of this array.
All flags can be accessed using the single (upper case) letter as well as the full name.
Examples
>>> y = np.array([[3, 1, 7], ... [2, 0, 0], ... [8, 5, 9]]) >>> y array([[3, 1, 7], [2, 0, 0], [8, 5, 9]]) >>> y.flags C_CONTIGUOUS : True F_CONTIGUOUS : False OWNDATA : True WRITEABLE : True ALIGNED : True WRITEBACKIFCOPY : False UPDATEIFCOPY : False >>> y.setflags(write=0, align=0) >>> y.flags C_CONTIGUOUS : True F_CONTIGUOUS : False OWNDATA : True WRITEABLE : False ALIGNED : False WRITEBACKIFCOPY : False UPDATEIFCOPY : False >>> y.setflags(uic=1) Traceback (most recent call last): File "<stdin>", line 1, in <module> ValueError: cannot set WRITEBACKIFCOPY flag to True
- shape
Tuple of array dimensions.
The shape property is usually used to get the current shape of an array, but may also be used to reshape the array in-place by assigning a tuple of array dimensions to it. As with numpy.reshape, one of the new shape dimensions can be -1, in which case its value is inferred from the size of the array and the remaining dimensions. Reshaping an array in-place will fail if a copy is required.
See also
numpy.reshape
similar function
ndarray.reshape
similar method
Examples
>>> x = np.array([1, 2, 3, 4]) >>> x.shape (4,) >>> y = np.zeros((2, 3, 4)) >>> y.shape (2, 3, 4) >>> y.shape = (3, 8) >>> y array([[ 0., 0., 0., 0., 0., 0., 0., 0.], [ 0., 0., 0., 0., 0., 0., 0., 0.], [ 0., 0., 0., 0., 0., 0., 0., 0.]]) >>> y.shape = (3, 6) Traceback (most recent call last): File "<stdin>", line 1, in <module> ValueError: total size of new array must be unchanged >>> np.zeros((4,2))[::2].shape = (-1,) Traceback (most recent call last): File "<stdin>", line 1, in <module> AttributeError: Incompatible shape for in-place modification. Use `.reshape()` to make a copy with the desired shape.
- size
Number of elements in the array.
Equal to
np.prod(a.shape)
, i.e., the product of the array’s dimensions.Notes
a.size returns a standard arbitrary precision Python integer. This may not be the case with other methods of obtaining the same value (like the suggested
np.prod(a.shape)
, which returns an instance ofnp.int_
), and may be relevant if the value is used further in calculations that may overflow a fixed size integer type.Examples
>>> x = np.zeros((3, 5, 2), dtype=np.complex128) >>> x.size 30 >>> np.prod(x.shape) 30
- sort(axis=- 1, kind=None, order=None)
Sort an array in-place. Refer to numpy.sort for full documentation.
- Parameters
- axisint, optional
Axis along which to sort. Default is -1, which means sort along the last axis.
- kind{‘quicksort’, ‘mergesort’, ‘heapsort’, ‘stable’}, optional
Sorting algorithm. The default is ‘quicksort’. Note that both ‘stable’ and ‘mergesort’ use timsort under the covers and, in general, the actual implementation will vary with datatype. The ‘mergesort’ option is retained for backwards compatibility.
Changed in version 1.15.0: The ‘stable’ option was added.
- orderstr or list of str, optional
When a is an array with fields defined, this argument specifies which fields to compare first, second, etc. A single field can be specified as a string, and not all fields need be specified, but unspecified fields will still be used, in the order in which they come up in the dtype, to break ties.
See also
numpy.sort
Return a sorted copy of an array.
numpy.argsort
Indirect sort.
numpy.lexsort
Indirect stable sort on multiple keys.
numpy.searchsorted
Find elements in sorted array.
numpy.partition
Partial sort.
Notes
See numpy.sort for notes on the different sorting algorithms.
Examples
>>> a = np.array([[1,4], [3,1]]) >>> a.sort(axis=1) >>> a array([[1, 4], [1, 3]]) >>> a.sort(axis=0) >>> a array([[1, 3], [1, 4]])
Use the order keyword to specify a field to use when sorting a structured array:
>>> a = np.array([('a', 2), ('c', 1)], dtype=[('x', 'S1'), ('y', int)]) >>> a.sort(order='y') >>> a array([(b'c', 1), (b'a', 2)], dtype=[('x', 'S1'), ('y', '<i8')])
- squeeze(axis=None)
Remove axes of length one from a.
Refer to numpy.squeeze for full documentation.
See also
numpy.squeeze
equivalent function
- std(axis=None, dtype=None, out=None, ddof=0, keepdims=False, *, where=True)
Returns the standard deviation of the array elements along given axis.
Refer to numpy.std for full documentation.
See also
numpy.std
equivalent function
- strides
Tuple of bytes to step in each dimension when traversing an array.
The byte offset of element
(i[0], i[1], ..., i[n])
in an array a is:offset = sum(np.array(i) * a.strides)
A more detailed explanation of strides can be found in the “ndarray.rst” file in the NumPy reference guide.
See also
numpy.lib.stride_tricks.as_strided
Notes
Imagine an array of 32-bit integers (each 4 bytes):
x = np.array([[0, 1, 2, 3, 4], [5, 6, 7, 8, 9]], dtype=np.int32)
This array is stored in memory as 40 bytes, one after the other (known as a contiguous block of memory). The strides of an array tell us how many bytes we have to skip in memory to move to the next position along a certain axis. For example, we have to skip 4 bytes (1 value) to move to the next column, but 20 bytes (5 values) to get to the same position in the next row. As such, the strides for the array x will be
(20, 4)
.Examples
>>> y = np.reshape(np.arange(2*3*4), (2,3,4)) >>> y array([[[ 0, 1, 2, 3], [ 4, 5, 6, 7], [ 8, 9, 10, 11]], [[12, 13, 14, 15], [16, 17, 18, 19], [20, 21, 22, 23]]]) >>> y.strides (48, 16, 4) >>> y[1,1,1] 17 >>> offset=sum(y.strides * np.array((1,1,1))) >>> offset/y.itemsize 17
>>> x = np.reshape(np.arange(5*6*7*8), (5,6,7,8)).transpose(2,3,1,0) >>> x.strides (32, 4, 224, 1344) >>> i = np.array([3,5,2,2]) >>> offset = sum(i * x.strides) >>> x[3,5,2,2] 813 >>> offset / x.itemsize 813
- sum(axis=None, dtype=None, out=None, keepdims=False, initial=0, where=True)
Return the sum of the array elements over the given axis.
Refer to numpy.sum for full documentation.
See also
numpy.sum
equivalent function
- swapaxes(axis1, axis2)
Return a view of the array with axis1 and axis2 interchanged.
Refer to numpy.swapaxes for full documentation.
See also
numpy.swapaxes
equivalent function
- take(indices, axis=None, out=None, mode='raise')
Return an array formed from the elements of a at the given indices.
Refer to numpy.take for full documentation.
See also
numpy.take
equivalent function
- tobytes(order='C')
Construct Python bytes containing the raw data bytes in the array.
Constructs Python bytes showing a copy of the raw contents of data memory. The bytes object is produced in C-order by default. This behavior is controlled by the
order
parameter.New in version 1.9.0.
- Parameters
- order{‘C’, ‘F’, ‘A’}, optional
Controls the memory layout of the bytes object. ‘C’ means C-order, ‘F’ means F-order, ‘A’ (short for Any) means ‘F’ if a is Fortran contiguous, ‘C’ otherwise. Default is ‘C’.
- Returns
- sbytes
Python bytes exhibiting a copy of a’s raw data.
Examples
>>> x = np.array([[0, 1], [2, 3]], dtype='<u2') >>> x.tobytes() b'\x00\x00\x01\x00\x02\x00\x03\x00' >>> x.tobytes('C') == x.tobytes() True >>> x.tobytes('F') b'\x00\x00\x02\x00\x01\x00\x03\x00'
- tofile(fid, sep='', format='%s')
Write array to a file as text or binary (default).
Data is always written in ‘C’ order, independent of the order of a. The data produced by this method can be recovered using the function fromfile().
- Parameters
- fidfile or str or Path
An open file object, or a string containing a filename.
Changed in version 1.17.0: pathlib.Path objects are now accepted.
- sepstr
Separator between array items for text output. If “” (empty), a binary file is written, equivalent to
file.write(a.tobytes())
.- formatstr
Format string for text file output. Each entry in the array is formatted to text by first converting it to the closest Python type, and then using “format” % item.
Notes
This is a convenience function for quick storage of array data. Information on endianness and precision is lost, so this method is not a good choice for files intended to archive data or transport data between machines with different endianness. Some of these problems can be overcome by outputting the data as text files, at the expense of speed and file size.
When fid is a file object, array contents are directly written to the file, bypassing the file object’s
write
method. As a result, tofile cannot be used with files objects supporting compression (e.g., GzipFile) or file-like objects that do not supportfileno()
(e.g., BytesIO).
- tolist()
Return the array as an
a.ndim
-levels deep nested list of Python scalars.Return a copy of the array data as a (nested) Python list. Data items are converted to the nearest compatible builtin Python type, via the ~numpy.ndarray.item function.
If
a.ndim
is 0, then since the depth of the nested list is 0, it will not be a list at all, but a simple Python scalar.- Parameters
- none
- Returns
- yobject, or list of object, or list of list of object, or …
The possibly nested list of array elements.
Notes
The array may be recreated via
a = np.array(a.tolist())
, although this may sometimes lose precision.Examples
For a 1D array,
a.tolist()
is almost the same aslist(a)
, except thattolist
changes numpy scalars to Python scalars:>>> a = np.uint32([1, 2]) >>> a_list = list(a) >>> a_list [1, 2] >>> type(a_list[0]) <class 'numpy.uint32'> >>> a_tolist = a.tolist() >>> a_tolist [1, 2] >>> type(a_tolist[0]) <class 'int'>
Additionally, for a 2D array,
tolist
applies recursively:>>> a = np.array([[1, 2], [3, 4]]) >>> list(a) [array([1, 2]), array([3, 4])] >>> a.tolist() [[1, 2], [3, 4]]
The base case for this recursion is a 0D array:
>>> a = np.array(1) >>> list(a) Traceback (most recent call last): ... TypeError: iteration over a 0-d array >>> a.tolist() 1
- tostring(order='C')
A compatibility alias for tobytes, with exactly the same behavior.
Despite its name, it returns bytes not strs.
Deprecated since version 1.19.0.
- trace(offset=0, axis1=0, axis2=1, dtype=None, out=None)
Return the sum along diagonals of the array.
Refer to numpy.trace for full documentation.
See also
numpy.trace
equivalent function
- transpose(*axes)
Returns a view of the array with axes transposed.
For a 1-D array this has no effect, as a transposed vector is simply the same vector. To convert a 1-D array into a 2D column vector, an additional dimension must be added. np.atleast2d(a).T achieves this, as does a[:, np.newaxis]. For a 2-D array, this is a standard matrix transpose. For an n-D array, if axes are given, their order indicates how the axes are permuted (see Examples). If axes are not provided and
a.shape = (i[0], i[1], ... i[n-2], i[n-1])
, thena.transpose().shape = (i[n-1], i[n-2], ... i[1], i[0])
.- Parameters
- axesNone, tuple of ints, or n ints
None or no argument: reverses the order of the axes.
tuple of ints: i in the j-th place in the tuple means a’s i-th axis becomes a.transpose()’s j-th axis.
n ints: same as an n-tuple of the same ints (this form is intended simply as a “convenience” alternative to the tuple form)
- Returns
- outndarray
View of a, with axes suitably permuted.
See also
transpose
Equivalent function
ndarray.T
Array property returning the array transposed.
ndarray.reshape
Give a new shape to an array without changing its data.
Examples
>>> a = np.array([[1, 2], [3, 4]]) >>> a array([[1, 2], [3, 4]]) >>> a.transpose() array([[1, 3], [2, 4]]) >>> a.transpose((1, 0)) array([[1, 3], [2, 4]]) >>> a.transpose(1, 0) array([[1, 3], [2, 4]])
- var(axis=None, dtype=None, out=None, ddof=0, keepdims=False, *, where=True)
Returns the variance of the array elements, along given axis.
Refer to numpy.var for full documentation.
See also
numpy.var
equivalent function
- view([dtype][, type])
New view of array with the same data.
Note
Passing None for
dtype
is different from omitting the parameter, since the former invokesdtype(None)
which is an alias fordtype('float_')
.- Parameters
- dtypedata-type or ndarray sub-class, optional
Data-type descriptor of the returned view, e.g., float32 or int16. Omitting it results in the view having the same data-type as a. This argument can also be specified as an ndarray sub-class, which then specifies the type of the returned object (this is equivalent to setting the
type
parameter).- typePython type, optional
Type of the returned view, e.g., ndarray or matrix. Again, omission of the parameter results in type preservation.
Notes
a.view()
is used two different ways:a.view(some_dtype)
ora.view(dtype=some_dtype)
constructs a view of the array’s memory with a different data-type. This can cause a reinterpretation of the bytes of memory.a.view(ndarray_subclass)
ora.view(type=ndarray_subclass)
just returns an instance of ndarray_subclass that looks at the same array (same shape, dtype, etc.) This does not cause a reinterpretation of the memory.For
a.view(some_dtype)
, ifsome_dtype
has a different number of bytes per entry than the previous dtype (for example, converting a regular array to a structured array), then the behavior of the view cannot be predicted just from the superficial appearance ofa
(shown byprint(a)
). It also depends on exactly howa
is stored in memory. Therefore ifa
is C-ordered versus fortran-ordered, versus defined as a slice or transpose, etc., the view may give different results.Examples
>>> x = np.array([(1, 2)], dtype=[('a', np.int8), ('b', np.int8)])
Viewing array data using a different type and dtype:
>>> y = x.view(dtype=np.int16, type=np.matrix) >>> y matrix([[513]], dtype=int16) >>> print(type(y)) <class 'numpy.matrix'>
Creating a view on a structured array so it can be used in calculations
>>> x = np.array([(1, 2),(3,4)], dtype=[('a', np.int8), ('b', np.int8)]) >>> xv = x.view(dtype=np.int8).reshape(-1,2) >>> xv array([[1, 2], [3, 4]], dtype=int8) >>> xv.mean(0) array([2., 3.])
Making changes to the view changes the underlying array
>>> xv[0,1] = 20 >>> x array([(1, 20), (3, 4)], dtype=[('a', 'i1'), ('b', 'i1')])
Using a view to convert an array to a recarray:
>>> z = x.view(np.recarray) >>> z.a array([1, 3], dtype=int8)
Views share data:
>>> x[0] = (9, 10) >>> z[0] (9, 10)
Views that change the dtype size (bytes per entry) should normally be avoided on arrays defined by slices, transposes, fortran-ordering, etc.:
>>> x = np.array([[1,2,3],[4,5,6]], dtype=np.int16) >>> y = x[:, 0:2] >>> y array([[1, 2], [4, 5]], dtype=int16) >>> y.view(dtype=[('width', np.int16), ('length', np.int16)]) Traceback (most recent call last): ... ValueError: To change to a dtype of a different size, the array must be C-contiguous >>> z = y.copy() >>> z.view(dtype=[('width', np.int16), ('length', np.int16)]) array([[(1, 2)], [(4, 5)]], dtype=[('width', '<i2'), ('length', '<i2')])
- class multimodal.datasets.data_sample.MultiModalSparseArray(*arg, **kwargs)
MultiModalArray inherit from numpy ndarray
- Parameters
- datacan be
- dictionary of multiview array with shape = (n_samples, n_features) for multi-view
for each view.
- {0: array([[]],
1: array([[]], …}
- numpy array like with shape = (n_samples, n_features) for multi-view
for each view.
- [[[…]],
[[…]], …]
- {array like} with (n_samples, nviews * n_features) with ‘views_ind’ diferent to ‘None’
for Multi-view input samples.
- views_indarray-like (default= None ) if None
[0, n_features//2, n_features]) is constructed (2 views) Paramater specifying how to extract the data views from X:
views_ind is a 1-D array of sorted integers, the entries indicate the limits of the slices used to extract the views, where view
n
is given byX[:, views_ind[n]:views_ind[n+1]]
.
- Attributes
- views_indlist of views’ indice (may be None)
n_views : int number of views
shapes_int: list of int numbers of feature for each views
keys : name of key, where data come from a dictionary
- :Example:
- >>> from multimodal.datasets.base import load_dict
- >>> from multimodal.tests.datasets.get_dataset_path import get_dataset_path
- >>> from multimodal.datasets.data_sample import DataSample
- >>> file = ‘input_x_dic.pkl’
- >>> data = load_dict(get_dataset_path(file))
- arcsin()
Element-wise arcsin.
See numpy.arcsin for more information.
- arcsinh()
Element-wise arcsinh.
See numpy.arcsinh for more information.
- arctan()
Element-wise arctan.
See numpy.arctan for more information.
- arctanh()
Element-wise arctanh.
See numpy.arctanh for more information.
- argmax(axis=None, out=None)
Return indices of maximum elements along an axis.
Implicit zero elements are also taken into account. If there are several maximum values, the index of the first occurrence is returned.
- Parameters
- axis{-2, -1, 0, 1, None}, optional
Axis along which the argmax is computed. If None (default), index of the maximum element in the flatten data is returned.
- outNone, optional
This argument is in the signature solely for NumPy compatibility reasons. Do not pass in anything except for the default value, as this argument is not used.
- Returns
- indnumpy.matrix or int
Indices of maximum elements. If matrix, its size along axis is 1.
- argmin(axis=None, out=None)
Return indices of minimum elements along an axis.
Implicit zero elements are also taken into account. If there are several minimum values, the index of the first occurrence is returned.
- Parameters
- axis{-2, -1, 0, 1, None}, optional
Axis along which the argmin is computed. If None (default), index of the minimum element in the flatten data is returned.
- outNone, optional
This argument is in the signature solely for NumPy compatibility reasons. Do not pass in anything except for the default value, as this argument is not used.
- Returns
- indnumpy.matrix or int
Indices of minimum elements. If matrix, its size along axis is 1.
- asformat(format, copy=False)
Return this matrix in the passed format.
- Parameters
- format{str, None}
The desired matrix format (“csr”, “csc”, “lil”, “dok”, “array”, …) or None for no conversion.
- copybool, optional
If True, the result is guaranteed to not share data with self.
- Returns
- AThis matrix in the passed format.
- asfptype()
Upcast matrix to a floating point format (if necessary)
- astype(dtype, casting='unsafe', copy=True)
Cast the matrix elements to a specified type.
- Parameters
- dtypestring or numpy dtype
Typecode or data-type to which to cast the data.
- casting{‘no’, ‘equiv’, ‘safe’, ‘same_kind’, ‘unsafe’}, optional
Controls what kind of data casting may occur. Defaults to ‘unsafe’ for backwards compatibility. ‘no’ means the data types should not be cast at all. ‘equiv’ means only byte-order changes are allowed. ‘safe’ means only casts which can preserve values are allowed. ‘same_kind’ means only safe casts or casts within a kind, like float64 to float32, are allowed. ‘unsafe’ means any data conversions may be done.
- copybool, optional
If copy is False, the result might share some memory with this matrix. If copy is True, it is guaranteed that the result and this matrix do not share any memory.
- ceil()
Element-wise ceil.
See numpy.ceil for more information.
- check_format(full_check=True)
check whether the matrix format is valid
- Parameters
- full_checkbool, optional
If True, rigorous check, O(N) operations. Otherwise basic check, O(1) operations (default True).
- conj(copy=True)
Element-wise complex conjugation.
If the matrix is of non-complex data type and copy is False, this method does nothing and the data is not copied.
- Parameters
- copybool, optional
If True, the result is guaranteed to not share data with self.
- Returns
- AThe element-wise complex conjugate.
- conjugate(copy=True)
Element-wise complex conjugation.
If the matrix is of non-complex data type and copy is False, this method does nothing and the data is not copied.
- Parameters
- copybool, optional
If True, the result is guaranteed to not share data with self.
- Returns
- AThe element-wise complex conjugate.
- copy()
Returns a copy of this matrix.
No data/indices will be shared between the returned value and current matrix.
- count_nonzero()
Number of non-zero entries, equivalent to
np.count_nonzero(a.toarray())
Unlike getnnz() and the nnz property, which return the number of stored entries (the length of the data attribute), this method counts the actual number of non-zero entries in data.
- deg2rad()
Element-wise deg2rad.
See numpy.deg2rad for more information.
- diagonal(k=0)
Returns the kth diagonal of the matrix.
- Parameters
- kint, optional
Which diagonal to get, corresponding to elements a[i, i+k]. Default: 0 (the main diagonal).
New in version 1.0.
See also
numpy.diagonal
Equivalent numpy function.
Examples
>>> from scipy.sparse import csr_matrix >>> A = csr_matrix([[1, 2, 0], [0, 0, 3], [4, 0, 5]]) >>> A.diagonal() array([1, 0, 5]) >>> A.diagonal(k=1) array([2, 3])
- dot(other)
Ordinary dot product
Examples
>>> import numpy as np >>> from scipy.sparse import csr_matrix >>> A = csr_matrix([[1, 2, 0], [0, 0, 3], [4, 0, 5]]) >>> v = np.array([1, 0, -1]) >>> A.dot(v) array([ 1, -3, -1], dtype=int64)
- eliminate_zeros()
Remove zero entries from the matrix
This is an in place operation.
- expm1()
Element-wise expm1.
See numpy.expm1 for more information.
- floor()
Element-wise floor.
See numpy.floor for more information.
- getH()
Return the Hermitian transpose of this matrix.
See also
numpy.matrix.getH
NumPy’s implementation of getH for matrices
- get_shape()
Get shape of a matrix.
- getcol(i)
Returns a copy of column i of the matrix, as a (m x 1) CSR matrix (column vector).
- getformat()
Format of a matrix representation as a string.
- getmaxprint()
Maximum number of elements to display when printed.
- getnnz(axis=None)
Number of stored values, including explicit zeros.
- Parameters
- axisNone, 0, or 1
Select between the number of values across the whole matrix, in each column, or in each row.
See also
count_nonzero
Number of non-zero entries
- getrow(i)
Returns a copy of row i of the matrix, as a (1 x n) CSR matrix (row vector).
- property has_canonical_format
Determine whether the matrix has sorted indices and no duplicates
- Returns
True: if the above applies
False: otherwise
has_canonical_format implies has_sorted_indices, so if the latter flag is False, so will the former be; if the former is found True, the latter flag is also set.
- property has_sorted_indices
Determine whether the matrix has sorted indices
- Returns
True: if the indices of the matrix are in sorted order
False: otherwise
- log1p()
Element-wise log1p.
See numpy.log1p for more information.
- max(axis=None, out=None)
Return the maximum of the matrix or maximum along an axis. This takes all elements into account, not just the non-zero ones.
- Parameters
- axis{-2, -1, 0, 1, None} optional
Axis along which the sum is computed. The default is to compute the maximum over all the matrix elements, returning a scalar (i.e., axis = None).
- outNone, optional
This argument is in the signature solely for NumPy compatibility reasons. Do not pass in anything except for the default value, as this argument is not used.
- Returns
- amaxcoo_matrix or scalar
Maximum of a. If axis is None, the result is a scalar value. If axis is given, the result is a sparse.coo_matrix of dimension
a.ndim - 1
.
See also
min
The minimum value of a sparse matrix along a given axis.
numpy.matrix.max
NumPy’s implementation of ‘max’ for matrices
- maximum(other)
Element-wise maximum between this and another matrix.
- mean(axis=None, dtype=None, out=None)
Compute the arithmetic mean along the specified axis.
Returns the average of the matrix elements. The average is taken over all elements in the matrix by default, otherwise over the specified axis. float64 intermediate and return values are used for integer inputs.
- Parameters
- axis{-2, -1, 0, 1, None} optional
Axis along which the mean is computed. The default is to compute the mean of all elements in the matrix (i.e., axis = None).
- dtypedata-type, optional
Type to use in computing the mean. For integer inputs, the default is float64; for floating point inputs, it is the same as the input dtype.
New in version 0.18.0.
- outnp.matrix, optional
Alternative output matrix in which to place the result. It must have the same shape as the expected output, but the type of the output values will be cast if necessary.
New in version 0.18.0.
- Returns
- mnp.matrix
See also
numpy.matrix.mean
NumPy’s implementation of ‘mean’ for matrices
- min(axis=None, out=None)
Return the minimum of the matrix or maximum along an axis. This takes all elements into account, not just the non-zero ones.
- Parameters
- axis{-2, -1, 0, 1, None} optional
Axis along which the sum is computed. The default is to compute the minimum over all the matrix elements, returning a scalar (i.e., axis = None).
- outNone, optional
This argument is in the signature solely for NumPy compatibility reasons. Do not pass in anything except for the default value, as this argument is not used.
- Returns
- amincoo_matrix or scalar
Minimum of a. If axis is None, the result is a scalar value. If axis is given, the result is a sparse.coo_matrix of dimension
a.ndim - 1
.
See also
max
The maximum value of a sparse matrix along a given axis.
numpy.matrix.min
NumPy’s implementation of ‘min’ for matrices
- minimum(other)
Element-wise minimum between this and another matrix.
- multiply(other)
Point-wise multiplication by another matrix, vector, or scalar.
- property nnz
Number of stored values, including explicit zeros.
See also
count_nonzero
Number of non-zero entries
- nonzero()
nonzero indices
Returns a tuple of arrays (row,col) containing the indices of the non-zero elements of the matrix.
Examples
>>> from scipy.sparse import csr_matrix >>> A = csr_matrix([[1,2,0],[0,0,3],[4,0,5]]) >>> A.nonzero() (array([0, 0, 1, 2, 2]), array([0, 1, 2, 0, 2]))
- power(n, dtype=None)
This function performs element-wise power.
- Parameters
- nn is a scalar
- dtypeIf dtype is not specified, the current dtype will be preserved.
- prune()
Remove empty space after all non-zero elements.
- rad2deg()
Element-wise rad2deg.
See numpy.rad2deg for more information.
- reshape(self, shape, order='C', copy=False)
Gives a new shape to a sparse matrix without changing its data.
- Parameters
- shapelength-2 tuple of ints
The new shape should be compatible with the original shape.
- order{‘C’, ‘F’}, optional
Read the elements using this index order. ‘C’ means to read and write the elements using C-like index order; e.g., read entire first row, then second row, etc. ‘F’ means to read and write the elements using Fortran-like index order; e.g., read entire first column, then second column, etc.
- copybool, optional
Indicates whether or not attributes of self should be copied whenever possible. The degree to which attributes are copied varies depending on the type of sparse matrix being used.
- Returns
- reshaped_matrixsparse matrix
A sparse matrix with the given shape, not necessarily of the same format as the current object.
See also
numpy.matrix.reshape
NumPy’s implementation of ‘reshape’ for matrices
- resize(*shape)
Resize the matrix in-place to dimensions given by
shape
Any elements that lie within the new shape will remain at the same indices, while non-zero elements lying outside the new shape are removed.
- Parameters
- shape(int, int)
number of rows and columns in the new matrix
Notes
The semantics are not identical to numpy.ndarray.resize or numpy.resize. Here, the same data will be maintained at each index before and after reshape, if that index is within the new bounds. In numpy, resizing maintains contiguity of the array, moving elements around in the logical matrix but not within a flattened representation.
We give no guarantees about whether the underlying data attributes (arrays, etc.) will be modified in place or replaced with new objects.
- rint()
Element-wise rint.
See numpy.rint for more information.
- set_shape(shape)
See reshape.
- setdiag(values, k=0)
Set diagonal or off-diagonal elements of the array.
- Parameters
- valuesarray_like
New values of the diagonal elements.
Values may have any length. If the diagonal is longer than values, then the remaining diagonal entries will not be set. If values are longer than the diagonal, then the remaining values are ignored.
If a scalar value is given, all of the diagonal is set to it.
- kint, optional
Which off-diagonal to set, corresponding to elements a[i,i+k]. Default: 0 (the main diagonal).
- property shape
Get shape of a matrix.
- sign()
Element-wise sign.
See numpy.sign for more information.
- sin()
Element-wise sin.
See numpy.sin for more information.
- sinh()
Element-wise sinh.
See numpy.sinh for more information.
- sort_indices()
Sort the indices of this matrix in place
- sorted_indices()
Return a copy of this matrix with sorted indices
- sqrt()
Element-wise sqrt.
See numpy.sqrt for more information.
- sum(axis=None, dtype=None, out=None)
Sum the matrix elements over a given axis.
- Parameters
- axis{-2, -1, 0, 1, None} optional
Axis along which the sum is computed. The default is to compute the sum of all the matrix elements, returning a scalar (i.e., axis = None).
- dtypedtype, optional
The type of the returned matrix and of the accumulator in which the elements are summed. The dtype of a is used by default unless a has an integer dtype of less precision than the default platform integer. In that case, if a is signed then the platform integer is used while if a is unsigned then an unsigned integer of the same precision as the platform integer is used.
New in version 0.18.0.
- outnp.matrix, optional
Alternative output matrix in which to place the result. It must have the same shape as the expected output, but the type of the output values will be cast if necessary.
New in version 0.18.0.
- Returns
- sum_along_axisnp.matrix
A matrix with the same shape as self, with the specified axis removed.
See also
numpy.matrix.sum
NumPy’s implementation of ‘sum’ for matrices
- sum_duplicates()
Eliminate duplicate matrix entries by adding them together
This is an in place operation.
- tan()
Element-wise tan.
See numpy.tan for more information.
- tanh()
Element-wise tanh.
See numpy.tanh for more information.
- toarray(order=None, out=None)
Return a dense ndarray representation of this matrix.
- Parameters
- order{‘C’, ‘F’}, optional
Whether to store multidimensional data in C (row-major) or Fortran (column-major) order in memory. The default is ‘None’, which provides no ordering guarantees. Cannot be specified in conjunction with the out argument.
- outndarray, 2-D, optional
If specified, uses this array as the output buffer instead of allocating a new array to return. The provided array must have the same shape and dtype as the sparse matrix on which you are calling the method. For most sparse types, out is required to be memory contiguous (either C or Fortran ordered).
- Returns
- arrndarray, 2-D
An array with the same shape and containing the same data represented by the sparse matrix, with the requested memory order. If out was passed, the same object is returned after being modified in-place to contain the appropriate values.
- tobsr(blocksize=None, copy=True)
Convert this matrix to Block Sparse Row format.
With copy=False, the data/indices may be shared between this matrix and the resultant bsr_matrix.
When blocksize=(R, C) is provided, it will be used for construction of the bsr_matrix.
- tocoo(copy=True)
Convert this matrix to COOrdinate format.
With copy=False, the data/indices may be shared between this matrix and the resultant coo_matrix.
- tocsc(copy=False)
Convert this matrix to Compressed Sparse Column format.
With copy=False, the data/indices may be shared between this matrix and the resultant csc_matrix.
- tocsr(copy=False)
Convert this matrix to Compressed Sparse Row format.
With copy=False, the data/indices may be shared between this matrix and the resultant csr_matrix.
- todense(order=None, out=None)
Return a dense matrix representation of this matrix.
- Parameters
- order{‘C’, ‘F’}, optional
Whether to store multi-dimensional data in C (row-major) or Fortran (column-major) order in memory. The default is ‘None’, which provides no ordering guarantees. Cannot be specified in conjunction with the out argument.
- outndarray, 2-D, optional
If specified, uses this array (or numpy.matrix) as the output buffer instead of allocating a new array to return. The provided array must have the same shape and dtype as the sparse matrix on which you are calling the method.
- Returns
- arrnumpy.matrix, 2-D
A NumPy matrix object with the same shape and containing the same data represented by the sparse matrix, with the requested memory order. If out was passed and was an array (rather than a numpy.matrix), it will be filled with the appropriate values and returned wrapped in a numpy.matrix object that shares the same memory.
- todia(copy=False)
Convert this matrix to sparse DIAgonal format.
With copy=False, the data/indices may be shared between this matrix and the resultant dia_matrix.
- todok(copy=False)
Convert this matrix to Dictionary Of Keys format.
With copy=False, the data/indices may be shared between this matrix and the resultant dok_matrix.
- tolil(copy=False)
Convert this matrix to List of Lists format.
With copy=False, the data/indices may be shared between this matrix and the resultant lil_matrix.
- trace(offset=0)
Returns the sum along diagonals of the sparse matrix.
- Parameters
- offsetint, optional
Which diagonal to get, corresponding to elements a[i, i+offset]. Default: 0 (the main diagonal).
- transpose(axes=None, copy=False)
Reverses the dimensions of the sparse matrix.
- Parameters
- axesNone, optional
This argument is in the signature solely for NumPy compatibility reasons. Do not pass in anything except for the default value.
- copybool, optional
Indicates whether or not attributes of self should be copied whenever possible. The degree to which attributes are copied varies depending on the type of sparse matrix being used.
- Returns
- pself with the dimensions reversed.
See also
numpy.matrix.transpose
NumPy’s implementation of ‘transpose’ for matrices
- trunc()
Element-wise trunc.
See numpy.trunc for more information.
Boosting
multimodal.boosting.mumbo
Multimodal Boosting
This module contains a MultiModal Boosting (MuMBo)
estimator for classification implemented in the MumboClassifier
class.
- class multimodal.boosting.mumbo.MumboClassifier(base_estimator=None, n_estimators=50, random_state=None, best_view_mode='edge')
It then iterates the process on the same dataset but where the weights of incorrectly classified instances are adjusted such that subsequent classifiers focus more on difficult cases. A MuMBo classifier.
A MuMBo classifier is a meta-estimator that implements a multimodal (or multi-view) boosting algorithm:
It fits a set of classifiers on the original dataset splitted into several views and retains the classifier obtained for the best view.
This class implements the MuMBo algorithm [1].
- Parameters
- base_estimatorobject, optional (default=DecisionTreeClassifier)
Base estimator from which the boosted ensemble is built. Support for sample weighting is required, as well as proper classes_ and n_classes_ attributes. The default is a DecisionTreeClassifie with parameter
max_depth=1
.- n_estimatorsinteger, optional (default=50)
Maximum number of estimators at which boosting is terminated.
- random_stateint, RandomState instance or None, optional (default=None)
If int, random_state is the seed used by the random number generator; If RandomState instance, random_state is the random number generator; If None, the random number generator is the RandomState instance used by np.random.
- best_view_mode{“edge”, “error”}, optional (default=”edge”)
Mode used to select the best view at each iteration:
if
best_view_mode == "edge"
, the best view is the view maximizing the edge value (variable δ (delta) in [1]),if
best_view_mode == "error"
, the best view is the view minimizing the classification error.
See also
sklearn.ensemble.AdaBoostClassifier
sklearn.ensemble.GradientBoostingClassifier
sklearn.tree.DecisionTreeClassifier
References
- 1(1,2)
Sokol Koço, “Tackling the uneven views problem with cooperation based ensemble learning methods”,
Examples
>>> from multimodal.boosting.mumbo import MumboClassifier >>> from sklearn.datasets import load_iris >>> X, y = load_iris(return_X_y=True) >>> views_ind = [0, 2, 4] # view 0: sepal data, view 1: petal data >>> clf = MumboClassifier(random_state=0) >>> clf.fit(X, y, views_ind) MumboClassifier(random_state=0) >>> print(clf.predict([[ 5., 3., 1., 1.]])) [1] >>> views_ind = [[0, 2], [1, 3]] # view 0: length data, view 1: width data >>> clf = MumboClassifier(random_state=0) >>> clf.fit(X, y, views_ind) MumboClassifier(random_state=0) >>> print(clf.predict([[ 5., 3., 1., 1.]])) [1]
>>> from sklearn.tree import DecisionTreeClassifier >>> base_estimator = DecisionTreeClassifier(max_depth=2) >>> clf = MumboClassifier(base_estimator=base_estimator, random_state=0) >>> clf.fit(X, y, views_ind) MumboClassifier(base_estimator=DecisionTreeClassifier(max_depth=2), random_state=0) >>> print(clf.predict([[ 5., 3., 1., 1.]])) [1]
- Attributes
- estimators_list of classifiers
Collection of fitted sub-estimators.
- classes_numpy.ndarray, shape = (n_classes,)
Classes labels.
- n_classes_int
Number of classes.
- estimator_weights_numpy.ndarray of floats, shape = (len(estimators
Weights for each estimator in the boosted ensemble.
- estimator_errors_array of floats
Empirical loss for each iteration.
- best_views_numpy.ndarray of integers, shape = (len(estimators_),)
Indices of the best view for each estimator in the boosted ensemble.
- property base_estimator_
Estimator used to grow the ensemble.
- decision_function(X)
Compute the decision function of X.
- Parameters
- X{ array-like, sparse matrix},
shape = (n_samples, n_views * n_features) Multi-view input samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR. maybe also MultimodalData
- Returns
- dec_funnumpy.ndarray, shape = (n_samples, k)
Decision function of the input samples. The order of outputs is the same of that of the classes_ attribute. Binary classification is a special cases with
k == 1
, otherwisek == n_classes
. For binary classification, values <=0 mean classification in the first class inclasses_
and values >0 mean classification in the second class inclasses_
.
- property estimator_
Estimator used to grow the ensemble.
- fit(X, y, views_ind=None)
Build a multimodal boosted classifier from the training set (X, y).
- Parameters
- Xdict dictionary with all views
or MultiModalData , MultiModalArray, MultiModalSparseArray or {array-like, sparse matrix}, shape = (n_samples, n_features) Training multi-view input samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR.
- yarray-like, shape = (n_samples,)
Target values (class labels).
- views_indarray-like (default=[0, n_features//2, n_features])
Paramater specifying how to extract the data views from X:
If views_ind is a 1-D array of sorted integers, the entries indicate the limits of the slices used to extract the views, where view
n
is given byX[:, views_ind[n]:views_ind[n+1]]
.With this convention each view is therefore a view (in the NumPy sense) of X and no copy of the data is done.
If views_ind is an array of arrays of integers, then each array of integers
views_ind[n]
specifies the indices of the viewn
, which is then given byX[:, views_ind[n]]
.With this convention each view creates therefore a partial copy of the data in X. This convention is thus more flexible but less efficient than the previous one.
- Returns
- selfobject
Returns self.
- get_params(deep=True)
Get parameters for this estimator.
- Parameters
- deepbool, default=True
If True, will return the parameters for this estimator and contained subobjects that are estimators.
- Returns
- paramsdict
Parameter names mapped to their values.
- predict(X)
Predict classes for X.
The predicted class of an input sample is computed as the weighted mean prediction of the classifiers in the ensemble.
- Parameters
- X{array-like, sparse matrix}, shape = (n_samples, n_features)
Multi-view input samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR.
- Returns
- ynumpy.ndarray, shape = (n_samples,)
Predicted classes.
- score(X, y)
Return the mean accuracy on the given test data and labels.
- Parameters
- X{array-like, sparse matrix} of shape = (n_samples, n_features)
Multi-view test samples. Sparse matrix can be CSC, CSR
- yarray-like, shape = (n_samples,)
True labels for X.
- Returns
- scorefloat
Mean accuracy of self.predict(X) wrt. y.
- set_params(**params)
Set the parameters of this estimator.
The method works on simple estimators as well as on nested objects (such as
Pipeline
). The latter have parameters of the form<component>__<parameter>
so that it’s possible to update each component of a nested object.- Parameters
- **paramsdict
Estimator parameters.
- Returns
- selfestimator instance
Estimator instance.
- staged_decision_function(X)
Compute decision function of X for each boosting iteration.
This method allows monitoring (i.e. determine error on testing set) after each boosting iteration.
- Parameters
- X{array-like, sparse matrix}, shape = (n_samples, n_features)
Multi-view input samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR. maybe also MultimodalData
- Returns
- dec_fungenerator of numpy.ndarrays, shape = (n_samples, k)
Decision function of the input samples. The order of outputs is the same of that of the classes_ attribute. Binary classification is a special cases with
k == 1
, otherwisek==n_classes
. For binary classification, values <=0 mean classification in the first class inclasses_
and values >0 mean classification in the second class inclasses_
.
- staged_predict(X)
Return staged predictions for X.
The predicted class of an input sample is computed as the weighted mean prediction of the classifiers in the ensemble.
This generator method yields the ensemble prediction after each iteration of boosting and therefore allows monitoring, such as to determine the prediction on a test set after each boost.
- Parameters
- X{array-like, sparse matrix} of shape = (n_samples, n_features)
Multi-view input samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR.
- Returns
- ygenerator of numpy.ndarrays, shape = (n_samples,)
Predicted classes.
- staged_score(X, y)
Return staged mean accuracy on the given test data and labels.
This generator method yields the ensemble score after each iteration of boosting and therefore allows monitoring, such as to determine the score on a test set after each boost.
- Parameters
- X{array-like, sparse matrix} of shape = (n_samples, n_features)
Multi-view test samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR.
- yarray-like, shape = (n_samples,)
True labels for X.
- Returns
- scoregenerator of floats
Mean accuracy of self.staged_predict(X) wrt. y.
multimodal.boosting.combo
This module contains a MultiConfusion MMatrix Bosting (CoMBo)
estimator for classification implemented in the MuComboClassifier
class.
- class multimodal.boosting.combo.MuComboClassifier(base_estimator=None, n_estimators=50, random_state=None)
It then iterates the process on the same dataset but where the weights of incorrectly classified instances are adjusted such that subsequent classifiers focus more on difficult cases. A MuCoMBo classifier.
A MuMBo classifier is a meta-estimator that implements a multimodal (or multi-view) boosting algorithm:
It fits a set of classifiers on the original dataset splitted into several views and retains the classifier obtained for the best view.
This class implements the MuMBo algorithm [1].
- Parameters
- base_estimatorobject, optional (default=DecisionTreeClassifier)
Base estimator from which the boosted ensemble is built. Support for sample weighting is required, as well as proper classes_ and n_classes_ attributes. The default is a DecisionTreeClassifier with parameter
max_depth=1
.- n_estimatorsinteger, optional (default=50)
Maximum number of estimators at which boosting is terminated.
- random_stateint, RandomState instance or None, optional (default=None)
If int, random_state is the seed used by the random number generator; If RandomState instance, random_state is the random number generator; If None, the random number generator is the RandomState instance used by np.random.
See also
sklearn.ensemble.AdaBoostClassifier
sklearn.ensemble.GradientBoostingClassifier
sklearn.tree.DecisionTreeClassifier
References
- 1
Koc{c}o, Sokol and Capponi, C{'e}cile A Boosting Approach to Multiview Classification with Cooperation, 2011,Proceedings of the 2011 European Conference on Machine Learning and Knowledge Discovery in Databases - Volume Part II, 209–228 Springer-Verlag https://link.springer.com/chapter/10.1007/978-3-642-23783-6_1
- 2
Sokol Koço, “Tackling the uneven views problem with cooperation based ensemble learning methods”, PhD Thesis, Aix-Marseille Université, 2013, http://www.theses.fr/en/2013AIXM4101.
Examples
>>> from multimodal.boosting.combo import MuComboClassifier >>> from sklearn.datasets import load_iris >>> X, y = load_iris(return_X_y=True) >>> views_ind = [0, 2, 4] # view 0: sepal data, view 1: petal data >>> clf = MuComboClassifier(random_state=0) >>> clf.fit(X, y, views_ind) MuComboClassifier(random_state=0) >>> print(clf.predict([[ 5., 3., 1., 1.]])) [0] >>> views_ind = [[0, 2], [1, 3]] # view 0: length data, view 1: width data >>> clf = MuComboClassifier(random_state=0) >>> clf.fit(X, y, views_ind) MuComboClassifier(random_state=0) >>> print(clf.predict([[ 5., 3., 1., 1.]])) [0]
>>> from sklearn.tree import DecisionTreeClassifier >>> base_estimator = DecisionTreeClassifier(max_depth=2) >>> clf = MuComboClassifier(base_estimator=base_estimator, random_state=1) >>> clf.fit(X, y, views_ind) MuComboClassifier(base_estimator=DecisionTreeClassifier(max_depth=2), random_state=1) >>> print(clf.predict([[ 5., 3., 1., 1.]])) [0]
- Attributes
- estimators_list of classifiers
Collection of fitted sub-estimators.
- classes_numpy.ndarray, shape = (n_classes,)
Classes labels.
- n_classes_int
Number of classes.
- n_views_int
Number of views
- estimator_weights_numpy.ndarray of floats, shape = (len(estimators_),)
Weights for each estimator in the boosted ensemble.
- estimator_errors_array of floats
Empirical loss for each iteration.
- best_views_numpy.ndarray of integers, shape = (len(estimators_),)
Indices of the best view for each estimator in the boosted ensemble.
- n_yi_numpy ndarray of int contains number of train sample for each classe shape (n_classes,)
- property base_estimator_
Estimator used to grow the ensemble.
- decision_function(X)
Compute the decision function of X.
- Parameters
- X{array-like, sparse matrix}, shape = (n_samples, n_features)
Multi-view input samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR.
- Returns
- dec_funnumpy.ndarray, shape = (n_view, n_samples, k)
Decision function of the input samples. The order of outputs is the same of that of the classes_ attribute. Binary classification is a special cases with
k == 1
, otherwisek == n_classes
. For binary classification, values <=0 mean classification in the first class inclasses_
and values >0 mean classification in the second class inclasses_
.
- property estimator_
Estimator used to grow the ensemble.
- fit(X, y, views_ind=None)
Build a multimodal boosted classifier from the training set (X, y).
- Parameters
- Xdict dictionary with all views
or MultiModalData , MultiModalArray, MultiModalSparseArray or {array-like, sparse matrix}, shape = (n_samples, n_features) Training multi-view input samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR.
- yarray-like, shape = (n_samples,)
Target values (class labels).
- views_indarray-like (default=[0, n_features//2, n_features])
Paramater specifying how to extract the data views from X:
If views_ind is a 1-D array of sorted integers, the entries indicate the limits of the slices used to extract the views, where view
n
is given byX[:, views_ind[n]:views_ind[n+1]]
.With this convention each view is therefore a view (in the NumPy sense) of X and no copy of the data is done.
If views_ind is an array of arrays of integers, then each array of integers
views_ind[n]
specifies the indices of the viewn
, which is then given byX[:, views_ind[n]]
.With this convention each view creates therefore a partial copy of the data in X. This convention is thus more flexible but less efficient than the previous one.
- Returns
- selfobject
Returns self.
- Raises
- ValueError estimator must support sample_weight
- ValueError where X and view_ind are not compatibles
- get_params(deep=True)
Get parameters for this estimator.
- Parameters
- deepbool, default=True
If True, will return the parameters for this estimator and contained subobjects that are estimators.
- Returns
- paramsdict
Parameter names mapped to their values.
- predict(X)
Predict classes for X.
The predicted class of an input sample is computed as the weighted mean prediction of the classifiers in the ensemble.
- Parameters
- X{array-like, sparse matrix}, shape = (n_samples, n_features)
Multi-view input samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR.
- Returns
- ynumpy.ndarray, shape = (n_samples,)
Predicted classes.
- Raises
- ValueError ‘X’ input matrix must be have the same total number of features
of ‘X’ fit data
- score(X, y)
Return the mean accuracy on the given test data and labels.
- Parameters
- X{array-like, sparse matrix} of shape = (n_samples, n_features)
Multi-view test samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR.
- yarray-like, shape = (n_samples,)
True labels for X.
- Returns
- scorefloat
Mean accuracy of self.predict(X) wrt. y.
- set_params(**params)
Set the parameters of this estimator.
The method works on simple estimators as well as on nested objects (such as
Pipeline
). The latter have parameters of the form<component>__<parameter>
so that it’s possible to update each component of a nested object.- Parameters
- **paramsdict
Estimator parameters.
- Returns
- selfestimator instance
Estimator instance.
- staged_decision_function(X)
Compute decision function of X for each boosting iteration.
This method allows monitoring (i.e. determine error on testing set) after each boosting iteration.
- Parameters
- X{array-like, sparse matrix}, shape = (n_samples, n_features)
Multi-view input samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR.
- Returns
- dec_fungenerator of numpy.ndarrays, shape = (n_samples, k)
Decision function of the input samples. The order of outputs is the same of that of the classes_ attribute. Binary classification is a special cases with
k == 1
, otherwisek==n_classes
. For binary classification, values <=0 mean classification in the first class inclasses_
and values >0 mean classification in the second class inclasses_
.
- staged_predict(X)
Return staged predictions for X.
The predicted class of an input sample is computed as the weighted mean prediction of the classifiers in the ensemble.
This generator method yields the ensemble prediction after each iteration of boosting and therefore allows monitoring, such as to determine the prediction on a test set after each boost.
- Parameters
- X{array-like, sparse matrix} of shape = (n_samples, n_features)
Multi-view input samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR.
- Returns
- ygenerator of numpy.ndarrays, shape = (n_samples,)
Predicted classes.
- staged_score(X, y)
Return staged mean accuracy on the given test data and labels.
This generator method yields the ensemble score after each iteration of boosting and therefore allows monitoring, such as to determine the score on a test set after each boost.
- Parameters
- X{array-like, sparse matrix} of shape = (n_samples, n_features)
Multi-view test samples. Sparse matrix can be CSC, CSR, COO, DOK, or LIL. COO, DOK and LIL are converted to CSR.
- yarray-like, shape = (n_samples,)
True labels for X.
- Returns
- scoregenerator of floats
Mean accuracy of self.staged_predict(X) wrt. y.
multimodal.boosting.boost
- class multimodal.boosting.boost.UBoosting
Abstract class MuComboClassifier and MumboClassifier should inherit from UBoosting for methods
Kernels
multimodal.kernels.mvml
- class multimodal.kernels.mvml.MVML(lmbda=0.1, eta=1, nystrom_param=1.0, kernel='linear', kernel_params=None, learn_A=1, learn_w=0, precision=0.0001, n_loops=6)
The MVML Classifier
- Parameters
- lmbdafloat regression_params lmbda (default = 0.1) for basic regularization
- etafloat regression_params eta (default = 1), first for basic regularization,
regularization of A (not necessary if A is not learned)
- kernellist of str (default: “precomputed”) if kernel is as input of fit function set kernel to
“precomputed” list or str indicate the metrics used for each kernels list of pairwise kernel function name (default : “precomputed”) if kernel is as input of fit function set kernel to “precomputed” example : [‘rbf’, ‘additive_chi2’, ‘linear’ ] for function defined in as PAIRWISE_KERNEL_FUNCTIONS
- kernel_paramslist of str defaultNone) list of dictionaries for parameters of kernel [{‘gamma’:50}
list of dict of corresponding kernels params KERNEL_PARAMS
- nystrom_param: value between 0 and 1 indicating level of nyström approximation; 1 = no approximation
- learn_Ainteger (default 1) choose if A is learned or not: 1 - yes (default);
2 - yes, sparse; 3 - no (MVML_Cov); 4 - no (MVML_I)
- learn_winteger (default 0) where learn w is needed
- precisionfloat (default1E-4) precision to stop algorithm
- n_loops(default 6) number of iterions
Examples
>>> from multimodal.kernels.mvml import MVML >>> from sklearn.datasets import load_iris >>> X, y = load_iris(return_X_y=True) >>> y[y>0] = 1 >>> views_ind = [0, 2, 4] # view 0: sepal data, view 1: petal data >>> clf = MVML() >>> clf.get_params() {'eta': 1, 'kernel': 'linear', 'kernel_params': None, 'learn_A': 1, 'learn_w': 0, 'lmbda': 0.1, 'n_loops': 6, 'nystrom_param': 1.0, 'precision': 0.0001} >>> clf.fit(X, y, views_ind) MVML() >>> print(clf.predict([[ 5., 3., 1., 1.]])) 0
- Attributes
- lmbdafloat regression_params lmbda (default = 0.1)
- etafloat regression_params eta (default = 1)
- regression_paramsarray/list of regression parameters
- kernellist or str indicate the metrics used for each kernels
list of pairwise kernel function name (default : “precomputed”) example : [‘rbf’, ‘additive_chi2’, ‘linear’ ] for function defined in as PAIRWISE_KERNEL_FUNCTIONS example kernel=[‘rbf’, ‘rbf’], for the first two views
- kernel_params: list of dict of corresponding kernels params KERNEL_PARAMS
- learn_A1 where Learn matrix A is needded
- learn_winteger where learn w is needed
- precisionfloat (default1E-4) precision to stop algorithm
- n_loopsnumber of itterions
- n_approxnumber of samples in approximation, equals n if no approx.
- classes_array like unique label for classes
- warning_messagedictionary with warning messages
- X_
metriclearning.datasets.data_sample.Metriclearn_array
array of input sample - K_
metriclearning.datasets.data_sample.Metriclearn_array
array of processed kernels - y_array-like, shape = (n_samples,)
Target values (class labels).
- regression_if the classifier is used as regression (defaultFalse)
- decision_function(X)
Compute the decision function of X.
- Parameters
- X{ array-like, sparse matrix},
shape = (n_samples, n_views * n_features) Multi-view input samples. maybe also MultimodalData
- Returns
- dec_funnumpy.ndarray, shape = (n_samples, )
Decision function of the input samples. For binary classification, values <=0 mean classification in the first class in
classes_
and values >0 mean classification in the second class inclasses_
.
- fit(X, y=None, views_ind=None)
Fit the MVML classifier
- Parameters
- X- Metriclearn_array {array-like, sparse matrix}, shape = (n_samples, n_features)
Training multi-view input samples. can be also Kernel where attibute ‘kernel’ is set to precompute “precomputed”
or - Dictionary of {array like} with shape = (n_samples, n_features) for multi-view
for each view.
Array of {array like} with shape = (n_samples, n_features) for multi-view for each view.
{array like} with (n_samples, nviews * n_features) with ‘views_ind’ diferent to ‘None’
- yarray-like, shape = (n_samples,)
Target values (class labels). array of length n_samples containing the classification/regression labels for training data
- views_indarray-like (default=[0, n_features//2, n_features])
Paramater specifying how to extract the data views from X:
views_ind is a 1-D array of sorted integers, the entries indicate the limits of the slices used to extract the views, where view
n
is given byX[:, views_ind[n]:views_ind[n+1]]
.With this convention each view is therefore a view (in the NumPy sense) of X and no copy of the data is done.
- Returns
- selfobject
Returns self.
- get_params(deep=True)
Get parameters for this estimator.
- Parameters
- deepbool, default=True
If True, will return the parameters for this estimator and contained subobjects that are estimators.
- Returns
- paramsdict
Parameter names mapped to their values.
- predict(X)
- Parameters
- Xdifferent formats are supported
Metriclearn_array {array-like, sparse matrix}, shape = (n_samples, n_features) Training multi-view input samples. can be also Kernel where attibute ‘kernel’ is set to precompute “precomputed”
Dictionary of {array like} with shape = (n_samples, n_features) for multi-view for each view.
Array of {array like} with shape = (n_samples, n_features) for multi-view for each view.
{array like} with (n_samples, nviews * n_features) with ‘views_ind’ diferent to ‘None’
- Returns
- ynumpy.ndarray, shape = (n_samples,)
Predicted classes.
- score(X, y)
Return the mean accuracy on the given test data and labels.
- Parameters
- X{array-like} of shape = (n_samples, n_features)
- yarray-like, shape = (n_samples,)
True labels for X.
- Returns
- scorefloat
Mean accuracy of self.predict(X) wrt. y.
- set_params(**params)
Set the parameters of this estimator.
The method works on simple estimators as well as on nested objects (such as
Pipeline
). The latter have parameters of the form<component>__<parameter>
so that it’s possible to update each component of a nested object.- Parameters
- **paramsdict
Estimator parameters.
- Returns
- selfestimator instance
Estimator instance.
multimodal.kernels.lpMKL
- class multimodal.kernels.lpMKL.MKL(lmbda, nystrom_param=1.0, kernel='linear', kernel_params=None, use_approx=True, precision=0.0001, n_loops=50)
MKL Classifier for multiview learning
- Parameters
- lmbdafloat coeficient for combined kernels
- nystrom_paramfloat (default1.0)
value between 0 and 1 indicating level of nyström approximation; 1 = no approximation
- kernellist of str (default: “precomputed”) if kernel is as input of fit function set kernel to
“precomputed” list or str indicate the metrics used for each kernels list of pairwise kernel function name (default : “precomputed”) if kernel is as input of fit function set kernel to “precomputed” example : [‘rbf’, ‘additive_chi2’, ‘linear’ ] for function defined in as PAIRWISE_KERNEL_FUNCTIONS
- kernel_paramslist of str defaultNone) list of dictionaries for parameters of kernel [{‘gamma’:50}
list of dict of corresponding kernels params KERNEL_PARAMS
- use_approx(defaultTrue) to use approximation of m_param < 1
- n_loops(default 50) number of iterions
- Attributes
- lmbdafloat coeficient for combined kernels
- m_paramfloat (default1.0)
value between 0 and 1 indicating level of nyström approximation; 1 = no approximation
- kernellist or str indicate the metrics used for each kernels
list of pairwise kernel function name (default : “precomputed”) example : [‘rbf’, ‘additive_chi2’, ‘linear’ ] for function defined in as PAIRWISE_KERNEL_FUNCTIONS example kernel=[‘rbf’, ‘rbf’], for the first two views
- kernel_params: list of dict of corresponding kernels params KERNEL_PARAMS
- precisionfloat (default1E-4) precision to stop algorithm
- n_loopsnumber of iterions
- classes_array like unique label for classes
- X_
metriclearning.datasets.data_sample.Metriclearn_array
array of input sample - K_
metriclearning.datasets.data_sample.Metriclearn_array
array of processed kernels - y_array-like, shape = (n_samples,)
Target values (class labels).
- Clearning solution that is learned in MKL
- weightslearned weight for combining the solutions of views, learned in
- decision_function(X)
Compute the decision function of X.
- Parameters
- Xdict dictionary with all views {array like} with shape = (n_samples, n_features) for multi-view
for each view. or MultiModalData , MultiModalArray or {array-like,}, shape = (n_samples, n_features) Training multi-view input samples. can be also Kernel where attibute ‘kernel’ is set to precompute “precomputed”
- Returns
- dec_funnumpy.ndarray, shape = (n_samples, )
Decision function of the input samples. For binary classification, values <=0 mean classification in the first class in
classes_
and values >0 mean classification in the second class inclasses_
.
- fit(X, y=None, views_ind=None)
- Parameters
- Xdifferent formats are supported
Metriclearn_array {array-like, sparse matrix}, shape = (n_samples, n_features) Training multi-view input samples. can be also Kernel where attibute ‘kernel’ is set to precompute “precomputed”
Dictionary of {array like} with shape = (n_samples, n_features) for multi-view for each view.
Array of {array like} with shape = (n_samples, n_features) for multi-view for each view.
{array like} with (n_samples, nviews * n_features) with ‘views_ind’ diferent to ‘None’
- yarray-like, shape = (n_samples,)
Target values (class labels). array of length n_samples containing the classification/regression labels for training data
- views_indarray-like (default=[0, n_features//2, n_features])
Paramater specifying how to extract the data views from X:
views_ind is a 1-D array of sorted integers, the entries indicate the limits of the slices used to extract the views, where view
n
is given byX[:, views_ind[n]:views_ind[n+1]]
.With this convention each view is therefore a view (in the NumPy sense) of X and no copy of the data is done.
- Returns
- selfobject
Returns self.
- get_params(deep=True)
Get parameters for this estimator.
- Parameters
- deepbool, default=True
If True, will return the parameters for this estimator and contained subobjects that are estimators.
- Returns
- paramsdict
Parameter names mapped to their values.
- learn_lpMKL()
function of lpMKL learning
- Returns
- return tuple (C, weights)
- lpMKL_predict(X, C, weights)
- Parameters
- Xarray-like test kernels precomputed array like
- Ccorresponding to Confusion learned matrix
- weightslearned weights
- Returns
- ynumpy.ndarray, shape = (n_samples,)
Predicted classes.
- predict(X)
- Parameters
- Xdict dictionary with all views {array like} with shape = (n_samples, n_features) for multi-view
for each view. or MultiModalData , MultiModalArray or {array-like,}, shape = (n_samples, n_features) Training multi-view input samples. can be also Kernel where attibute ‘kernel’ is set to precompute “precomputed”
- views_indarray-like (default=[0, n_features//2, n_features])
Paramater specifying how to extract the data views from X:
views_ind is a 1-D array of sorted integers, the entries indicate the limits of the slices used to extract the views, where view
n
is given byX[:, views_ind[n]:views_ind[n+1]]
.With this convention each view is therefore a view (in the NumPy sense) of X and no copy of the data is done.
- Returns
- ynumpy.ndarray, shape = (n_samples,)
Predicted classes.
- score(X, y)
Return the mean accuracy on the given test data and labels.
- Parameters
- Xdict dictionary with all views {array like} with shape = (n_samples, n_features) for multi-view
for each view. or MultiModalData , MultiModalArray or {array-like,}, shape = (n_samples, n_features) Training multi-view input samples. can be also Kernel where attibute ‘kernel’ is set to precompute “precomputed”
- yarray-like, shape = (n_samples,)
True labels for X.
- Returns
- scorefloat
Mean accuracy of self.predict(X) wrt. y.
- set_params(**params)
Set the parameters of this estimator.
The method works on simple estimators as well as on nested objects (such as
Pipeline
). The latter have parameters of the form<component>__<parameter>
so that it’s possible to update each component of a nested object.- Parameters
- **paramsdict
Estimator parameters.
- Returns
- selfestimator instance
Estimator instance.
multimodal.kernels.mkernel
- class multimodal.kernels.mkernel.MKernel
Abstract class MKL and MVML should inherit from for methods of transform kernel to/from data.
- Attributes
- W_sqrootinv_dictdict of nyström approximation kernel
in the case of nystrom approximation the a dictonary of reduced kernel is calculated
- kernel_paramslist of dict of corresponding kernels
params KERNEL_PARAMS