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:

data : dict

kwargs : others arguments

Attributes

clear() → None. Remove all items from D.

copy() → a shallow copy of D

data: MultiModalArray

fromkeys($type, iterable, value=None, /): Returns a new dict with keys from iterable and values equal to value.

get(k[, d]) → D[k] if k in D, else d. d defaults to None.

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(k[, d]) → v, remove specified key and return the corresponding value.: If key is not found, d is returned if given, otherwise KeyError is raised

popitem() → (k, v), remove and return some (key, value) pair as a: 2-tuple; but raise KeyError if D is empty.

setdefault(k[, d]) → D.get(k,d), also set D[k]=d if k not in D

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

MultiModalArray inherit from numpy ndarray

Parameters:

data : can 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_ind : array-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 by X[:, views_ind[n]:views_ind[n+1]].

Attributes

views_ind	(list of views’ indice (may be None))
n_views	(int 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

Same as self.transpose(), except that self is returned if self.ndim < 2.

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)

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)

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 of a.

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='quicksort', 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:

dtype : str 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.

subok : bool, optional

If True, then sub-classes will be passed-through (default), otherwise the returned array will be forced to be a base-class array.

copy : bool, 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_t : ndarray

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

Starting in NumPy 1.9, astype method now returns an error if the string dtype to cast to is not long enough in ‘safe’ casting mode to hold the max value of integer/float array that is being casted. Previously the casting was allowed even if the result was truncated.

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)

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.

Parameters:

inplace : bool, optional

If True, swap bytes in-place, default is False.

Returns:

out : ndarray

The byteswapped array. If inplace is True, this is a view to self.

Examples

>>> A = np.array([1, 256, 8755], dtype=np.int16)
>>> map(hex, A)
['0x1', '0x100', '0x2233']
>>> A.byteswap(True)
array([  256,     1, 13090], dtype=int16)
>>> map(hex, A)
['0x100', '0x1', '0x3322']

Arrays of strings are not swapped

>>> A = np.array(['ceg', 'fac'])
>>> A.byteswap()
array(['ceg', 'fac'],
      dtype='|S3')

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)

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 :func:numpy.copy are very similar, but have different default values for their order= arguments.)

See also

numpy.copy, numpy.copyto

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:

c : Python 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):

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].
shape (c_intp*self.ndim): 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 c_int, c_long, or 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.
strides (c_intp*self.ndim): 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.
data_as(obj): Return the data pointer cast to a particular c-types object. For example, calling self._as_parameter_ is equivalent to self.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)).
shape_as(obj): Return the shape tuple as an array of some other c-types type. For example: self.shape_as(ctypes.c_short).
strides_as(obj): Return the strides tuple as an array of some other c-types type. For example: self.strides_as(ctypes.c_longlong).

Be careful using the ctypes attribute - especially on temporary arrays or arrays constructed on the fly. For example, calling (a+b).ctypes.data_as(ctypes.c_void_p) returns a pointer to memory that is invalid because the array created as (a+b) is deallocated before the next Python statement. You can avoid this problem using either c=a+b or ct=(a+b).ctypes. In the latter case, ct will hold a reference to the array until ct is deleted or re-assigned.

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
array([[0, 1],
       [2, 3]])
>>> x.ctypes.data
30439712
>>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_long))
<ctypes.LP_c_long object at 0x01F01300>
>>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_long)).contents
c_long(0)
>>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_longlong)).contents
c_longlong(4294967296L)
>>> x.ctypes.shape
<numpy.core._internal.c_long_Array_2 object at 0x01FFD580>
>>> x.ctypes.shape_as(ctypes.c_long)
<numpy.core._internal.c_long_Array_2 object at 0x01FCE620>
>>> x.ctypes.strides
<numpy.core._internal.c_long_Array_2 object at 0x01FCE620>
>>> x.ctypes.strides_as(ctypes.c_longlong)
<numpy.core._internal.c_longlong_Array_2 object at 0x01F01300>

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:	d : numpy 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:

file : str

A string naming the dump file.

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:

value : scalar

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 in a.flags.writeable). Short flag names are only supported in dictionary access.

Only the 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.
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 if arr.shape[dim] == 1 or the array has no elements. It does not generally hold that self.strides[-1] == self.itemsize for C-style contiguous arrays or self.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.
UPDATEIFCOPY (U)	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)
<type '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:

y : ndarray

A copy of the input array, flattened to one dimension.

See also

ravel: Return a flattened array.
flat: A 1-D flat iterator over the array.

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:

dtype : str or dtype

The data type of the view. The dtype size of the view can not be larger than that of the array itself.

offset : int

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:

*args : Arguments (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:

z : Standard 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

>>> x = np.random.randint(9, size=(3, 3))
>>> x
array([[3, 1, 7],
       [2, 8, 3],
       [8, 5, 3]])
>>> x.item(3)
2
>>> x.item(7)
5
>>> x.item((0, 1))
1
>>> x.item((2, 2))
3

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 than a[args] = item. The item should be a scalar value and args must select a single item in the array a.

Parameters:

*args : Arguments

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

>>> x = np.random.randint(9, size=(3, 3))
>>> x
array([[3, 1, 7],
       [2, 8, 3],
       [8, 5, 3]])
>>> x.itemset(4, 0)
>>> x.itemset((2, 2), 9)
>>> x
array([[3, 1, 7],
       [2, 0, 3],
       [8, 5, 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)

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)

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)

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_order : string, 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

{‘<’, ‘L’} - little endian

{‘>’, ‘B’} - big endian

{‘=’, ‘N’} - native order

{‘|’, ‘I’} - ignore (no change to byte order)

The default value (‘S’) results in swapping the current byte order. The code does a case-insensitive check on the first letter of new_order for the alternatives above. For example, any of ‘B’ or ‘b’ or ‘biggish’ are valid to specify big-endian.

Returns:

new_arr : array

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 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:

kth : int 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 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.

axis : int, optional

Axis along which to sort. Default is -1, which means sort along the last axis.

kind : {‘introselect’}, optional

Selection algorithm. Default is ‘introselect’.

order : str 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.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))
array([1, 2, 3, 4])

prod(axis=None, dtype=None, out=None, keepdims=False)

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)

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

resize(new_shape, refcheck=True)

Change shape and size of array in-place.

Parameters:

new_shape : tuple of ints, or n ints

Shape of resized array.

refcheck : bool, 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 has been 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:

val : object

Value to be placed in field.

dtype : dtype object

Data-type of the field in which to place val.

offset : int, optional

The number of bytes into the field at which to place val.

Returns:

None

See also

getfield

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]])
>>> x
array([[  1.00000000e+000,   1.48219694e-323,   1.48219694e-323],
       [  1.48219694e-323,   1.00000000e+000,   1.48219694e-323],
       [  1.48219694e-323,   1.48219694e-323,   1.00000000e+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, 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 UPDATEIFCOPY flag 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:

write : bool, optional

Describes whether or not a can be written to.

align : bool, optional

Describes whether or not a is aligned properly for its type.

uic : bool, 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 6 Boolean flags in use, only three of which can be changed by the user: 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) this array is a copy of some other array (referenced by .base). When this array is deallocated, the base array will be updated with the contents of this array.

All flags can be accessed using their first (upper case) letter as well as the full name.

Examples

>>> 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
  UPDATEIFCOPY : False
>>> y.setflags(write=0, align=0)
>>> y.flags
  C_CONTIGUOUS : True
  F_CONTIGUOUS : False
  OWNDATA : True
  WRITEABLE : False
  ALIGNED : False
  UPDATEIFCOPY : False
>>> y.setflags(uic=1)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: cannot set UPDATEIFCOPY flag to True

shape

Tuple of array dimensions.

Notes

May be used to “reshape” the array, as long as this would not require a change in the total number of elements

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

size

Number of elements in the array.

Equivalent to np.prod(a.shape), i.e., the product of the array’s dimensions.

Examples

>>> x = np.zeros((3, 5, 2), dtype=np.complex128)
>>> x.size
30
>>> np.prod(x.shape)
30

sort(axis=-1, kind='quicksort', order=None)

Sort an array, in-place.

Parameters:

axis : int, optional

Axis along which to sort. Default is -1, which means sort along the last axis.

kind : {‘quicksort’, ‘mergesort’, ‘heapsort’}, optional

Sorting algorithm. Default is ‘quicksort’.

order : str 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.
argsort: Indirect sort.
lexsort: Indirect stable sort on multiple keys.
searchsorted: Find elements in sorted array.
partition: Partial sort.

Notes

See 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([('c', 1), ('a', 2)],
      dtype=[('x', '|S1'), ('y', '<i4')])

squeeze(axis=None)

Remove single-dimensional entries from the shape of 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)

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)

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 can be produced in either ‘C’ or ‘Fortran’, or ‘Any’ order (the default is ‘C’-order). ‘Any’ order means C-order unless the F_CONTIGUOUS flag in the array is set, in which case it means ‘Fortran’ order.

New in version 1.9.0.

Parameters:

order : {‘C’, ‘F’, None}, optional

Order of the data for multidimensional arrays: C, Fortran, or the same as for the original array.

Returns:

s : bytes

Python bytes exhibiting a copy of a’s raw data.

Examples

>>> x = np.array([[0, 1], [2, 3]])
>>> x.tobytes()
b'\x00\x00\x00\x00\x01\x00\x00\x00\x02\x00\x00\x00\x03\x00\x00\x00'
>>> x.tobytes('C') == x.tobytes()
True
>>> x.tobytes('F')
b'\x00\x00\x00\x00\x02\x00\x00\x00\x01\x00\x00\x00\x03\x00\x00\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:

fid : file or str

An open file object, or a string containing a filename.

sep : str

Separator between array items for text output. If “” (empty), a binary file is written, equivalent to file.write(a.tobytes()).

format : str

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.

tolist()

Return the array as a (possibly nested) list.

Return a copy of the array data as a (nested) Python list. Data items are converted to the nearest compatible Python type.

Parameters:

none

Returns:

y : list

The possibly nested list of array elements.

Notes

The array may be recreated, a = np.array(a.tolist()).

Examples

>>> a = np.array([1, 2])
>>> a.tolist()
[1, 2]
>>> a = np.array([[1, 2], [3, 4]])
>>> list(a)
[array([1, 2]), array([3, 4])]
>>> a.tolist()
[[1, 2], [3, 4]]

tostring(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 can be produced in either ‘C’ or ‘Fortran’, or ‘Any’ order (the default is ‘C’-order). ‘Any’ order means C-order unless the F_CONTIGUOUS flag in the array is set, in which case it means ‘Fortran’ order.

This function is a compatibility alias for tobytes. Despite its name it returns bytes not strings.

Parameters:

order : {‘C’, ‘F’, None}, optional

Order of the data for multidimensional arrays: C, Fortran, or the same as for the original array.

Returns:

s : bytes

Python bytes exhibiting a copy of a’s raw data.

Examples

>>> x = np.array([[0, 1], [2, 3]])
>>> x.tobytes()
b'\x00\x00\x00\x00\x01\x00\x00\x00\x02\x00\x00\x00\x03\x00\x00\x00'
>>> x.tobytes('C') == x.tobytes()
True
>>> x.tobytes('F')
b'\x00\x00\x00\x00\x02\x00\x00\x00\x01\x00\x00\x00\x03\x00\x00\x00'

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. (To change between column and row vectors, first cast the 1-D array into a matrix object.) For a 2-D array, this is the usual 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]), then a.transpose().shape = (i[n-1], i[n-2], ... i[1], i[0]).

Parameters:

axes : None, 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:

out : ndarray

View of a, with axes suitably permuted.

See also

ndarray.T: Array property returning the array transposed.

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)

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=None, type=None)

New view of array with the same data.

Parameters:

dtype : data-type or ndarray sub-class, optional

Data-type descriptor of the returned view, e.g., float32 or int16. The default, None, 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).

type : Python type, optional

Type of the returned view, e.g., ndarray or matrix. Again, the default None results in type preservation.

Notes

a.view() is used two different ways:

a.view(some_dtype) or a.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) or a.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), if some_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 of a (shown by print(a)). It also depends on exactly how a is stored in memory. Therefore if a 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.matrixlib.defmatrix.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
>>> print(x)
[(1, 20) (3, 4)]

Using a view to convert an array to a recarray:

>>> z = x.view(np.recarray)
>>> z.a
array([1], 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):
  File "<stdin>", line 1, in <module>
ValueError: new type not compatible with array.
>>> 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:

data : can 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_ind : array-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 by X[:, views_ind[n]:views_ind[n+1]].

Attributes

views_ind	(list 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 minimum 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.

out : None, 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:

ind : np.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.

out : None, 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:

ind : np.matrix or int

Indices of minimum elements. If matrix, its size along axis is 1.

asformat(format)

Return this matrix in a given sparse format

Parameters:

format : {string, None}

desired sparse matrix format

None for no format conversion

“csr” for csr_matrix format

“csc” for csc_matrix format

“lil” for lil_matrix format

“dok” for dok_matrix format and so on

asfptype(): Upcast matrix to a floating point format (if necessary)

astype(t)

Cast the matrix elements to a specified type.

The data will be copied.

Parameters:

t : string or numpy dtype

Typecode or data-type to which to cast the data.

ceil()

Element-wise ceil.

See numpy.ceil for more information.

check_format(full_check=True)

check whether the matrix format is valid

Parameters:

full_check : bool, optional

If True, rigorous check, O(N) operations. Otherwise basic check, O(1) operations (default True).

conj()

Element-wise complex conjugation.

If the matrix is of non-complex data type, then this method does nothing and the data is not copied.

conjugate()

Element-wise complex conjugation.

If the matrix is of non-complex data type, then this method does nothing and the data is not copied.

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(): Returns the main diagonal of the matrix

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

np.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:

axis : None, 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).

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.

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).

out : None, 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:

amax : coo_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.
np.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).

dtype : data-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.

out : np.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.

Returns:

m : np.matrix

See also

np.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).

out : None, 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:

amin : coo_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.
np.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.

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:

n : n is a scalar

dtype : If 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(shape, order='C')

Gives a new shape to a sparse matrix without changing its data.

Parameters:

shape : length-2 tuple of ints

The new shape should be compatible with the original shape.

order : ‘C’, 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:

reshaped_matrix : self with the new dimensions of shape

See also

np.matrix.reshape: NumPy’s implementation of ‘reshape’ for matrices

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:

values : array_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 if longer than the diagonal, then the remaining values are ignored.

If a scalar value is given, all of the diagonal is set to it.

k : int, optional

Which off-diagonal to set, corresponding to elements a[i,i+k]. Default: 0 (the main diagonal).

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).

dtype : dtype, 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.

out : np.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.

Returns:

sum_along_axis : np.matrix

A matrix with the same shape as self, with the specified axis removed.

See also

np.matrix.sum: NumPy’s implementation of ‘sum’ for matrices

sum_duplicates()

Eliminate duplicate matrix entries by adding them together

The 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): See the docstring for spmatrix.toarray.

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’, indicating the NumPy default of C-ordered. Cannot be specified in conjunction with the out argument.

out : ndarray, 2-dimensional, 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:

arr : numpy.matrix, 2-dimensional

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 LInked List format.

With copy=False, the data/indices may be shared between this matrix and the resultant lil_matrix.

transpose(axes=None, copy=False)

Reverses the dimensions of the sparse matrix.

Parameters:

axes : None, optional

This argument is in the signature solely for NumPy compatibility reasons. Do not pass in anything except for the default value.

copy : bool, 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:

p : self with the dimensions reversed.

See also

np.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 [R11].

Parameters:

base_estimator : object, 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_estimators : integer, optional (default=50)

Maximum number of estimators at which boosting is terminated.

random_state : int, 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 [R11]),

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

[R11]

(1, 2, 3) 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.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.

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_fun : numpy.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, otherwise k == n_classes. For binary classification, values <=0 mean classification in the first class in classes_ and values >0 mean classification in the second class in classes_.

fit(X, y, views_ind=None)

Build a multimodal boosted classifier from the training set (X, y).

Parameters:

X : dict 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.

y : array-like, shape = (n_samples,)

Target values (class labels).

views_ind : array-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 by X[:, 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 view n, which is then given by X[:, 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:

self : object

Returns self.

get_params(deep=True)

Get parameters for this estimator.

Parameters:

deep : bool, default=True

If True, will return the parameters for this estimator and contained subobjects that are estimators.

Returns:

params : dict

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:

y : numpy.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

y : array-like, shape = (n_samples,)

True labels for X.

Returns:

score : float

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:

**params : dict

Estimator parameters.

Returns:

self : estimator 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_fun : generator 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, otherwise k==n_classes. For binary classification, values <=0 mean classification in the first class in classes_ and values >0 mean classification in the second class in classes_.

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:

y : generator 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.

y : array-like, shape = (n_samples,)

True labels for X.

Returns:

score : generator 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 [R34].

Parameters:

base_estimator : object, 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_estimators : integer, optional (default=50)

Maximum number of estimators at which boosting is terminated.

random_state : int, 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

[R34]

(1, 2) 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

[R44]

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,))

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_fun : numpy.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, otherwise k == n_classes. For binary classification, values <=0 mean classification in the first class in classes_ and values >0 mean classification in the second class in classes_.

fit(X, y, views_ind=None)

Build a multimodal boosted classifier from the training set (X, y).

Parameters:

X : dict 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.

y : array-like, shape = (n_samples,)

Target values (class labels).

views_ind : array-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 by X[:, 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 view n, which is then given by X[:, 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:

self : object

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:

deep : bool, default=True

If True, will return the parameters for this estimator and contained subobjects that are estimators.

Returns:

params : dict

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:

y : numpy.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.

y : array-like, shape = (n_samples,)

True labels for X.

Returns:

score : float

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:

**params : dict

Estimator parameters.

Returns:

self : estimator 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_fun : generator 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, otherwise k==n_classes. For binary classification, values <=0 mean classification in the first class in classes_ and values >0 mean classification in the second class in classes_.

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:

y : generator 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.

y : array-like, shape = (n_samples,)

True labels for X.

Returns:

score : generator 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:

lmbda : float regression_params lmbda (default = 0.1) for basic regularization

eta : float regression_params eta (default = 1), first for basic regularization,

regularization of A (not necessary if A is not learned)

kernel : list 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_params : list of str default

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_A : integer (default 1) choose if A is learned or not: 1 - yes (default);

2 - yes, sparse; 3 - no (MVML_Cov); 4 - no (MVML_I)

learn_w : integer (default 0) where learn w is needed

precision : float (default

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

lmbda	(float regression_params lmbda (default = 0.1))
eta	(float regression_params eta (default = 1))
regression_params	(array/list of regression parameters)
kernel	(list 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_A	( 1 where Learn matrix A is needded)
learn_w	(integer where learn w is needed)
precision	(float (default)
n_loops	(number of itterions)
n_approx	(number of samples in approximation, equals n if no approx.)
classes_	(array like unique label for classes)
warning_message	(dictionary 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 (default)

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_fun : numpy.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 in classes_.

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’

y : array-like, shape = (n_samples,)

Target values (class labels). array of length n_samples containing the classification/regression labels for training data

views_ind : array-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 by X[:, 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:

self : object

Returns self.

get_params(deep=True)

Get parameters for this estimator.

Parameters:

deep : bool, default=True

If True, will return the parameters for this estimator and contained subobjects that are estimators.

Returns:

params : dict

Parameter names mapped to their values.

predict(X)

Parameters:

X : different 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:

y : numpy.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)

y : array-like, shape = (n_samples,)

True labels for X.

Returns:

score : float

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:

**params : dict

Estimator parameters.

Returns:

self : estimator 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:

lmbda : float coeficient for combined kernels

nystrom_param : float (default

value between 0 and 1 indicating level of nyström approximation; 1 = no approximation

kernel : list 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_params : list of str default

list of dict of corresponding kernels params KERNEL_PARAMS

use_approx : (default

n_loops : (default 50) number of iterions

Attributes

lmbda	(float coeficient for combined kernels)
m_param	(float (default) value between 0 and 1 indicating level of nyström approximation; 1 = no approximation
kernel	(list 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
precision	(float (default)
n_loops	(number 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).
C	(learning solution that is learned in MKL)
weights	(learned weight for combining the solutions of views, learned in)

decision_function(X)

Compute the decision function of X.

Parameters:

X : dict 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_fun : numpy.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 in classes_.

fit(X, y=None, views_ind=None)

Parameters:

X : different 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’

y : array-like, shape = (n_samples,)

Target values (class labels). array of length n_samples containing the classification/regression labels for training data

views_ind : array-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 by X[:, 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:

self : object

Returns self.

get_params(deep=True)

Get parameters for this estimator.

Parameters:

deep : bool, default=True

If True, will return the parameters for this estimator and contained subobjects that are estimators.

Returns:

params : dict

Parameter names mapped to their values.

learn_lpMKL()

function of lpMKL learning

Returns:	return tuple (C, weights)

lpMKL_predict(X, C, weights)

Parameters:

X : array-like test kernels precomputed array like

C : corresponding to Confusion learned matrix

weights : learned weights

Returns:

y : numpy.ndarray, shape = (n_samples,)

Predicted classes.

predict(X)

Parameters:

X : dict 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_ind : array-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 by X[:, 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:

y : numpy.ndarray, shape = (n_samples,)

Predicted classes.

score(X, y)

Return the mean accuracy on the given test data and labels.

Parameters:

X : dict 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”

y : array-like, shape = (n_samples,)

True labels for X.

Returns:

score : float

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:

**params : dict

Estimator parameters.

Returns:

self : estimator 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_dict	(dict of nyström approximation kernel) in the case of nystrom approximation the a dictonary of reduced kernel is calculated
kernel_params	(list of dict of corresponding kernels) params KERNEL_PARAMS