From 100b147b8b58ee7a954c12232fa500331ad4e065 Mon Sep 17 00:00:00 2001
From: Dominique Benielli <dominique.benielli@lis-lab.fr>
Date: Fri, 10 Jan 2020 20:21:48 +0100
Subject: [PATCH] test still not ok but improve
---
multimodal/boosting/boost.py | 79 +--
multimodal/boosting/cumbo.py | 29 +-
multimodal/boosting/mumbo.py | 32 +-
multimodal/datasets/__init__.py | 4 +-
.../__pycache__/__init__.cpython-36.pyc | Bin 319 -> 304 bytes
.../datasets/__pycache__/base.cpython-36.pyc | Bin 1258 -> 1249 bytes
.../__pycache__/data_sample.cpython-36.pyc | Bin 10701 -> 16022 bytes
multimodal/datasets/base.py | 2 +-
multimodal/datasets/data_sample.py | 490 ++++++++++++++----
multimodal/kernels/__init__.py | 2 +-
multimodal/kernels/lpMKL.py | 2 +-
multimodal/kernels/mvml.py | 8 +-
multimodal/tests/test.py | 224 ++++++++
multimodal/tests/test_cumbo.py | 4 +-
multimodal/tests/test_data_sample.py | 14 +-
multimodal/tests/test_mkl.py | 12 +-
multimodal/tests/test_mumbo.py | 31 +-
multimodal/tests/test_mvml.py | 18 +-
18 files changed, 732 insertions(+), 219 deletions(-)
create mode 100644 multimodal/tests/test.py
diff --git a/multimodal/boosting/boost.py b/multimodal/boosting/boost.py
index 7de84ee..7b039e6 100644
--- a/multimodal/boosting/boost.py
+++ b/multimodal/boosting/boost.py
@@ -1,11 +1,14 @@
import numpy as np
+import scipy.sparse as sp
from abc import ABCMeta
from sklearn.utils import check_array, check_X_y, check_random_state
from sklearn.tree import DecisionTreeClassifier
from sklearn.tree.tree import BaseDecisionTree
from sklearn.tree._tree import DTYPE
from sklearn.ensemble.forest import BaseForest
-from multimodal.datasets.data_sample import DataSample, MultiModalArray
+from multimodal.datasets.data_sample import DataSample
+from multimodal.datasets.data_sample import MultiModalData, MultiModalArray, MultiModalSparseArray
+
class UBoosting(metaclass=ABCMeta):
"""
@@ -22,60 +25,32 @@ class UBoosting(metaclass=ABCMeta):
else:
check_array(X, accept_sparse=['csr', 'csc'])
- if X.shape[1] != self.n_features_:
- raise ValueError("X doesn't contain the right number of features.")
- return X
+ if X.ndim < 2:
+ mes = "Reshape your data"
+ raise ValueError(mes)
+ if X.ndim > 1:
+ if X.shape[1] != self.n_features_:
+ mes = "Reshape your data"
+ raise ValueError("Number of features of the model must "
+ "match the input. Model n_features is %s and "
+ "input n_features is %s " % (self.n_features_, X.shape[1]))
- def _validate_views_ind(self, views_ind, n_features):
- """Ensure proper format for views_ind and return number of views."""
- views_ind = np.array(views_ind)
- if np.issubdtype(views_ind.dtype, np.integer) and views_ind.ndim == 1:
- if np.any(views_ind[:-1] >= views_ind[1:]):
- raise ValueError("Values in views_ind must be sorted.")
- if views_ind[0] < 0 or views_ind[-1] > n_features:
- raise ValueError("Values in views_ind are not in a correct "
- + "range for the provided data.")
- self.view_mode_ = "slices"
- n_views = views_ind.shape[0]-1
- else:
- if views_ind.ndim == 1:
- if not views_ind.dtype == np.object:
- raise ValueError("The format of views_ind is not "
- + "supported.")
- for ind, val in enumerate(views_ind):
- views_ind[ind] = np.array(val)
- if not np.issubdtype(views_ind[ind].dtype, np.integer):
- raise ValueError("Values in views_ind must be "
- + "integers.")
- if views_ind[ind].min() < 0 \
- or views_ind[ind].max() >= n_features:
- raise ValueError("Values in views_ind are not in a "
- + "correct range for the provided "
- + "data.")
- elif views_ind.ndim == 2:
- if not np.issubdtype(views_ind.dtype, np.integer):
- raise ValueError("Values in views_ind must be integers.")
- if views_ind.min() < 0 or views_ind.max() >= n_features:
- raise ValueError("Values in views_ind are not in a "
- + "correct range for the provided data.")
- else:
- raise ValueError("The format of views_ind is not supported.")
- self.view_mode_ = "indices"
- n_views = views_ind.shape[0]
- return (views_ind, n_views)
+ #
+ # raise ValueError(mes)
+ return X
def _global_X_transform(self, X, views_ind=None):
X_ = None
- if isinstance(X, np.ndarray) and X.ndim == 1:
- X_= MultiModalArray(X, views_ind)
- elif isinstance(X, dict):
- X_= MultiModalArray(X)
- elif isinstance(X, np.ndarray) and X.ndim > 1:
+ if isinstance(X, sp.spmatrix):
+ X_ = MultiModalSparseArray(X, views_ind)
+ else:
X_ = MultiModalArray(X, views_ind)
- if not isinstance(X_, MultiModalArray):
- raise TypeError("Input format is not reconized")
- if hasattr(self, "X_"):
- if not self.X_.viexs_ind == views_ind:
- raise ValueError("Input format (viewd, features) for fit and predict must be the same")
- return X_
\ No newline at end of file
+ if isinstance(X, MultiModalData):
+ X_ = X
+ if not isinstance(X_, MultiModalData):
+ try:
+ X_ = np.asarray(X)
+ except Exception as e:
+ raise TypeError('Reshape your data')
+ return X_
diff --git a/multimodal/boosting/cumbo.py b/multimodal/boosting/cumbo.py
index 3b22496..0f928df 100644
--- a/multimodal/boosting/cumbo.py
+++ b/multimodal/boosting/cumbo.py
@@ -78,13 +78,6 @@ class MuCumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
If None, the random number generator is the RandomState instance used
by `np.random`.
- best_view_mode : {"edge", "error"}, optional (default="edge")
- Mode used to select the best view at each iteration:
-
- - if ``best_view_mode == "edge"``, the best view is the view maximizing
- the edge value (variable δ (*delta*) in [1]_),
- - if ``best_view_mode == "error"``, the best view is the view
- minimizing the classification error.
Attributes
----------
@@ -120,15 +113,13 @@ class MuCumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
>>> views_ind = [0, 2, 4] # view 0: sepal data, view 1: petal data
>>> clf = MuCumboClassifier(random_state=0)
>>> clf.fit(X, y, views_ind) # doctest: +NORMALIZE_WHITESPACE
- MumboClassifier(base_estimator=None, best_view_mode='edge',
- n_estimators=50, 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 = MuCumboClassifier(random_state=0)
>>> clf.fit(X, y, views_ind) # doctest: +NORMALIZE_WHITESPACE
- MumboClassifier(base_estimator=None, best_view_mode='edge',
- n_estimators=50, random_state=0)
+
>>> print(clf.predict([[ 5., 3., 1., 1.]]))
[1]
@@ -136,13 +127,7 @@ class MuCumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
>>> base_estimator = DecisionTreeClassifier(max_depth=2)
>>> clf = MuCumboClassifier(base_estimator=base_estimator, random_state=0)
>>> clf.fit(X, y, views_ind) # doctest: +NORMALIZE_WHITESPACE
- MumboClassifier(base_estimator=DecisionTreeClassifier(class_weight=None,
- criterion='gini', max_depth=2, max_features=None,
- max_leaf_nodes=None, min_impurity_decrease=0.0,
- min_impurity_split=None, min_samples_leaf=1, min_samples_split=2,
- min_weight_fraction_leaf=0.0, presort=False, random_state=None,
- splitter='best'),
- best_view_mode='edge', n_estimators=50, random_state=0)
+
>>> print(clf.predict([[ 5., 3., 1., 1.]]))
[1]
@@ -176,7 +161,6 @@ class MuCumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
base_estimator=base_estimator,
n_estimators=n_estimators)
self.random_state = random_state
- # self.best_view_mode = self._validate_best_view_mode(best_view_mode)
def _validate_estimator(self):
"""Check the estimator and set the base_estimator_ attribute."""
@@ -527,7 +511,8 @@ class MuCumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
``classes_``.
"""
check_is_fitted(self, ("estimators_", "estimator_weights_alpha_","n_views_",
- "estimator_weights_beta_", "n_classes_"))
+ "estimator_weights_beta_", "n_classes_", "X_"))
+ X = self._global_X_transform(X, views_ind=self.X_.views_ind)
X = self._validate_X_predict(X)
n_samples = X.shape[0]
@@ -581,6 +566,7 @@ class MuCumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
"""
check_is_fitted(self, ("estimators_", "estimator_weights_alpha_","n_views_",
"estimator_weights_beta_", "n_classes_"))
+ X = self._global_X_transform(X, views_ind=self.X_.views_ind)
X = self._validate_X_predict(X)
n_samples = X.shape[0]
@@ -605,7 +591,7 @@ class MuCumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
else:
yield np.array(dec_func)
- def predict(self, X, views_ind=None):
+ def predict(self, X):
"""Predict classes for X.
The predicted class of an input sample is computed as the weighted mean
@@ -628,7 +614,6 @@ class MuCumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
ValueError 'X' input matrix must be have the same total number of features
of 'X' fit data
"""
- X = self._global_X_transform(X, views_ind=views_ind)
pred = self.decision_function(X)
if self.n_classes_ == 2:
diff --git a/multimodal/boosting/mumbo.py b/multimodal/boosting/mumbo.py
index c2fb588..f2c522d 100644
--- a/multimodal/boosting/mumbo.py
+++ b/multimodal/boosting/mumbo.py
@@ -32,6 +32,7 @@ estimator for classification implemented in the ``MumboClassifier`` class.
# structure, notations and behavior where possible.
import numpy as np
+
from sklearn.base import ClassifierMixin
from sklearn.ensemble import BaseEnsemble
from sklearn.ensemble.forest import BaseForest
@@ -343,22 +344,23 @@ class MumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
else:
dtype = None
accept_sparse = ['csr', 'csc']
- if views_ind is None:
- if X.shape[1] > 1:
- views_ind = np.array([0, X.shape[1]//2, X.shape[1]])
- else:
- views_ind = np.array([0, X.shape[1]])
- self.views_ind_, n_views = self._validate_views_ind(views_ind,
- X.shape[1])
- self.X_ = self._global_X_transform(X, views_ind=self.views_ind_)
+ # if views_ind is None:
+ # if X.shape[1] > 1:
+ # views_ind = np.array([0, X.shape[1]//2, X.shape[1]])
+ # elif X.shape[1]==1:
+ # views_ind = np.array([0, X.shape[1]])
+ # else:
+ # views_ind = np.array([0])
+ self.X_ = self._global_X_transform(X, views_ind=views_ind)
+ views_ind_, n_views = self.X_._validate_views_ind(self.X_.views_ind,
+ self.X_.shape[1])
check_X_y(self.X_, y, accept_sparse=accept_sparse, dtype=dtype)
check_classification_targets(y)
self._validate_estimator()
self.classes_, y = np.unique(y, return_inverse=True)
self.n_classes_ = len(self.classes_)
- self.n_features_ = X.shape[1]
-
+ self.n_features_ = self.X_.shape[1]
if self.n_classes_ == 1:
# This case would lead to division by 0 when computing the cost
# matrix so it needs special handling (but it is an obvious case as
@@ -458,7 +460,8 @@ class MumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
``classes_``.
"""
check_is_fitted(self, ("estimators_", "estimator_weights_",
- "best_views_", "n_classes_", "views_ind_"))
+ "best_views_", "n_classes_", "X_"))
+ X = self._global_X_transform(X, views_ind=self.X_.views_ind)
X = self._validate_X_predict(X)
n_samples = X.shape[0]
@@ -504,7 +507,8 @@ class MumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
``classes_``.
"""
check_is_fitted(self, ("estimators_", "estimator_weights_",
- "n_classes_", "views_ind_"))
+ "n_classes_", "X_"))
+ X = self._global_X_transform(X, views_ind=self.X_.views_ind)
X = self._validate_X_predict(X)
n_samples = X.shape[0]
@@ -542,6 +546,7 @@ class MumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
y : numpy.ndarray, shape = (n_samples,)
Predicted classes.
"""
+
pred = self.decision_function(X)
if self.n_classes_ == 2:
@@ -588,8 +593,7 @@ class MumboClassifier(BaseEnsemble, ClassifierMixin, UBoosting):
----------
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.
+ Sparse matrix can be CSC, CSR
y : array-like, shape = (n_samples,)
True labels for X.
diff --git a/multimodal/datasets/__init__.py b/multimodal/datasets/__init__.py
index 3bffe20..8b45df1 100644
--- a/multimodal/datasets/__init__.py
+++ b/multimodal/datasets/__init__.py
@@ -1,2 +1,2 @@
-from metriclearning.datasets.base import *
-from metriclearning.datasets.data_sample import DataSample, Metriclearn_array
\ No newline at end of file
+from multimodal.datasets.base import *
+from multimodal.datasets.data_sample import DataSample, MultiModalArray
\ No newline at end of file
diff --git a/multimodal/datasets/__pycache__/__init__.cpython-36.pyc b/multimodal/datasets/__pycache__/__init__.cpython-36.pyc
index 71e809afc1d58e7dd739fc5d6790139d3a7a4230..78203c1c83371c7711d386af2cea6da83faec7cb 100644
GIT binary patch
delta 121
zcmdnbw1J7!n3tF9e}_n1-bBt2OMc(doRUo6{FKBT$D*RdN<U4OD2ZH<ST0CRFD0=g
wu{gD)ST8BDIQ5o1LT2Iu9ma@>XO$F-lQXk3OLP%3IjM<7dHM+cWKl*f07iT)od5s;
delta 136
zcmdnMw4aI7n3tDJF{muIVj^dVv!HKkNl|8UPHJLNUVLIvQDUW^CQFn|E<!XjFI_Js
wu_UoLwWL@tDX}>9mNKgB#A!N=u@es~spXdDlw_7=rk111>!WffvoUG`0P*-Td;kCd
diff --git a/multimodal/datasets/__pycache__/base.cpython-36.pyc b/multimodal/datasets/__pycache__/base.cpython-36.pyc
index c952741223ea079326c463e7176cb64f88b719a4..5d08b5d131a62ac27ac9450f9ba3386c212792bd 100644
GIT binary patch
delta 79
zcmaFG`H+*-n3tDJzEdRb!A4G9M#hlI=8Q^;#mSl3nI*cpr8y;;x%nxHIXS6`MS1!N
S{^okdn@rsD2!YAzEX@EByBpa6
delta 88
zcmaFJ`HGX%n3tD}Z~MX6CmT6+85tudn=>k@<(B4@WR_*7mglCH6lEsoq$U>SW#*;p
UqjEP_G2Ue2S4I__9LLfO0CRO9o&W#<
diff --git a/multimodal/datasets/__pycache__/data_sample.cpython-36.pyc b/multimodal/datasets/__pycache__/data_sample.cpython-36.pyc
index 0a201db1c4aaece161b8408837f5bd47c3c60df1..aad5511d634098208d0e9aa1cb41329d923a21d6 100644
GIT binary patch
literal 16022
zcmXr!<>h*`Pa<wz0t3Ti1|-1Fz`)?Zz`#)4$H2gl!Vtxf!kEI8!<fqy#l#3=Gv_cz
zv7|7jFz2vlu|=_gS?pOHQJn4!DJ&_hEet8FsZ1$sz06Tu?hGmHDI6^fDICpAQQYnf
zDV!->Eet7Kshn9nshnB7sobet&1_M8?hGm1DLgF<DLknxS^TM7%`8y@?hGlsDSRyq
zDSWAd&CF3kDa^qPn*51tnHU&CGBS%5a`RJ4b5a$O^Ycm)GxLfSN-|OvToOwXgA;QL
zK;k)x#l;GVc_|9MsU=03$vLTsMS1awMMa5~3dtZjt`KY*QuC4%3yMo~5=&Bx6%rMS
zA%>|JD<tRV7Ubup=9Lt4<t6!m4bDq)tx(7-%}q)zD%Rmbw+3R1LS|k@YEfp1LRwLN
zE=X}fr9xf`#2FyR!JVI|V3eYel30=mljbVVEXe@-syHLDAXPylFCOBWVjYFN__EB@
z^7u>zErq=JwA93s(xTMj_)JYME&~N9Fa*&?Ai|i-777$VG=ku&fC2>&jUa}+ILMnI
zjPNR$#l;0mBajGFK#4?9Oz42aIxWAns3b!nB{R7sGe0k}C^NNKAuYcMnh<oj3W`$G
zGAncxic8bdGAqEzAT2SuB)>?ZJR`Lz6=X)GLShjpUFH^)mZYX|y#!?~O~za7j!w?L
zsU?Yi$shq3W`<=p4F(2=RE8+V6ox3KRAx|?Yhh?+jABV)PGM<bh+<7)O<`+ch+<1&
z4`$HhxW(sNnp2YLo1c=H14{JCOi%;a7#J8>85kIxK^eG(k%6IvA&aqwp_#E3#A2#p
z$YL&G$YN<`D(b0W$YM=l%w{U;DPaStNMXumDGDoL$YQTyNMTM9X=RdRr~zkq5G$Lt
znXxFWgd>HenX!qnh9QMjk|BjPo1r+bh9N~%lA(q%o296%gd;@^B%i`j3$lw7tcp>B
zp_X|9L*6@>jT0D)B%m(g%4RBFQ^JuV4$@VzuY?=q5*LQpm>8y7mReS@>sgx_YZz)+
zY@j}<VUdLUqWDD#M~VbUFGz(XLl#dn*i{o4i@Bg8yv-oDu@-GB;Yg7JsVv+U&X5-Z
zb_-KB(*(w%loEz4kZYkKGl8+#u7o2+0Hn4C*~F+ChAgOQ6BvtfN;p!aL2B|iN;p!4
zK&%vDgs*FvYZz*nYnakN#g<H!ooilkX;G>|K~a7|Y7r<&<tCPZC<RbXE{@O4O93aQ
zqSTVoqC8MRk(64bke>#T)_cjwz`*bl6v=*7%3+B)rK!aVnRyUNu-@F#;u3|VRE6UF
zqLS1Uy&{nMDot!EL5VXjzXYT`Q6V|Ms3<kLM4?KiC^0WR73?5TCMhV&FUw3xg_KBo
zx7dnvGLut_tK>mtW?Fs`$n7w<fvwLhRsfk<#Z_EdPylg4l?*o9VeYKr%*-oEO;0T<
z)~iy)rVz!MRq`;0;&5gaE2!wyt5U^g6wFS9jVc6;gBYesz_$p$-eS+pO96$@%m4rX
z|1aWTU|^_n#pV<gUs2f+#CqO=QioB?Yt#uLO}<-9c?Gvv!FB2_uFT@%(xjA<%7WBe
ztPqMF8t=DQ@=`K$Z!zbj=G|gW%&WY`1r7+;qN4nwTdd$p_7+=yQdVkm$t}**ywcp%
zqQsKaTg<tcdAFEz6Du^?ZgIjh>n$!s#e0hjBpjccpOPAXiycxo7vEye%uBh&T$Y%#
zlA%bJfq~&yw0=f@ZmNDter{%7W?^ZnenC-wR%%JHzN24|u1{uJYO#KCa%OgBiEb{a
zY|jOi?ci36K7y|gN}0tVO&|ueL7-Pqc}p@rEwiY&B)%*$Co?6nBo*91fD|jBngG<E
zV_@cDl48_i<Y1IwWMX7vWMLFy;$h?gvxOM>7)78g0Y)aUxDX=;qYNVxBL@=)BMT!J
zW0eMKouh}-N_cGqDzHE`61X<<U|?XVVaQ?t*EtF`3@J=0Os$}Dm$8OXk|B*rk|7vW
z-=K)rfNC$s6lQS-NrqM?h?*3pU<OT=Do59flA^@q5>S;3E(qa;S9)eyYMw%7UP@{O
zhy(`(r2Pdd+CTwUoLZs~p;rWI>G~D1FfcG^G8IWOFfbH}gW0!Oic@paZgFJhrGPbq
z;}1f}F)%RP;*C#**bSBi`KuV@O9mz`Mlq%;8T4R6R-6o~@IiGoNCUj0o&>9?8Jihv
z8Nt;NBfO&St6_vz)O{svHH^)SNEJ0ByrO0VwZpMg)QsRN7*bI)f-7oJHIGzL*D%C$
z)-Yyqfz`pP>JKnGComRC)j-_Boy}Cdt^`q4^MKrfrK)CxSJjLtRrM=)RSgm6g;&))
zH4qWLW`-K3Y}TS3@Tz(TxT*&EA5vAPp@z)_#$pF}Rm}*hoIy>J35>-tHH=wM(<U$$
z<zcI;k^NoET*Fwy5YJ!3T*DAAP{UZmoCZo{6j$5ev{eL(=7FfTi$G~?w9bZ<g3uNU
zqRs~AesEO{Ca6`Z5#MMSsr9vB5htjq<pL4hAc6-(@PY_F5Wx>31VDr!hya(%LLinf
zh!6o0q98&HM1TtHA}LU*%~qrU7mxt4K%zy`AeIb>0M)@rWw<bws?r}+Rl0ymXC~CD
zl8uC_l8upzk&Tg!u}TTO9>A=d{WRHcam2^xCFZ8a$KT?L2elzeb3kmK`1r!o#2ly$
zPjN|NNoI0xYDq?ZN|8LsJ4zrz8AKpF24;b3s3K4;17d*dB@QMY4Gwm&m?q;b&fL_J
zMDPd*_KqkhSYaK}6qXdG6y_GjDCQK#U<OUrTOtTQ1s5b16{mXUrRDp<D>YEWf?^X~
z-`g=VFr+ioGN&-svVi(=>7ZUHYYk%-Q!QHxQ!RT6b1g>+a~2Dz8Y$WYu5ehgS&BB*
zFr_eMa~364@|7@TvDGj&GuASNGZe04WB|bu_7ql7Un`QKa9spLo)1_xN71?(#w?B$
z7D<K_kQuW|7_vB17~x`UQVcM5Iy1<;JP}kC>0n!G7;9N;7_zu(m|&`#nPQl0nQGZ;
zIck`)SW{TCnTmeZaAa|(us1Un{Yv3T;cQ_j;i+M&VFCpesQPVYasm018EgV0*gXsr
z7>hxbD9n8f6Bvt>N_bPaK>n!Vs9}ibt6@T_e%NXl;swF29PTOw=ls0llA==Z5JP?%
z=3o}Mr{@PQ>cO>5l_InOs!*JfUz(Ew>J;W=7MCdG7b)ZwfE&r+3h5R*s85%fm!1sj
zDuU8HD9wP{LcyS10Lnq>3^fd~0x?XrjI~UlAdz60$W+J@3~A1SvyvwBEhaq!P*c|s
zL>PeuAp|J({o=9#Sq&OGvzrPs71AE8GQ^gO^y1?)^D;}~<MnKE@{<#DitY3ex~pU}
z^9o8!AOmC&k0qsoyPT=H1tpc5Lf|sx7CUrk?G{UMW>xAfrs9HI9GS(%;FMYfDqL={
zCl-Ul;1*{{WkD*q?Oy~gT)@Q<C}-W`DlSPZDk%mHzujW4$}A{C6holW<`x&Ys{pE1
zN+2Z<C%ELf#R6(s-r@lF1v2weip)VxMOJV#^cF{6YI!_J9<1XQdr3xWc4}oYq(cG9
z^{J3hgXdsSu>c8<Dp0_gfg*^B1>6p2V`O1sVG?2FU}R$CV`O6#VB}!rV^m@iVPauo
zV-#W(VU%G~VB`Q7LTrpxvUmeqlhIF;rAP%-j;MkNH4tG3BEU{T5TN)h(qv#@r~tVK
z6p<WEU_MGI0ul%HHHty$2VRPR$`4RE!Vtxh!U8TwMA04P2pX78oWRDwzy%t!L`Z`r
z@Qk8zae*}iBo-y+rk11@6@$6DZ~&181-*imLULlBLQ*PN1;`dK)<q2VRe}awz@05n
zBr1RtiUKGdD1f6}!4_>07B=##sQ~VKfOYDEw4&Mowgxnan3|ZJ0q&~kp)061u!2~p
z5gi>HtAj4BV2EE@PfxEFRgW%c2&n)x6j_v5sgRSIok|~%C`3m`>*?uX^Me9z8ITTi
z_g6!LALR5}h}%&{z(K)Wte~ZUGytv%QKk--0u4*4E2Lzmr52^;l_-?tE2#VB=cTHn
zTB87oIdF0V$D9JBoPmzC>v4eqIQ|qAz%d8P_9+U`*w6(zPC+9jH7&6;r^HqPWPyUF
zLS`D6g<%uOoM;2k=shgq=<6FHb7M6XGC?y9&=Nm2MM1*|Vyh-r!@!9?u_U!fp|~J5
zIWw&?GcR2sBflKvTWF^f+&c!ResHKjy1@}vkX!+B5D4o+ECB@>$N`B8hPp10<Os?T
zkRe3{=qO4tWSl%TuLLwUj^-@zC|F_%co<wECo?y*q!?rZsM`&yZBmOBN{drd(47Z%
z8G4$B45x#n6cQ5h5)wcTgLLeZD$#6BNQj8G(n0tkIxp4=q!Aof(Rtd2v9SpWdI&E&
zmXs7_CY6??Ldpto2%wc0ATf{;;2^P50M%k3kAQ@V)j@*_naQaN3L3eIm7w|+6z`e{
z^I;XR6)4O=`4D+17oiMXrhwd7g0LVnuLP<A9B|Nd0xn(Pg-0>M2v8lOV5I<RC4!un
zom!~_oreHNd2)VkDkR1ekqS93h<+>A3h?9>G%eWK*+G<oOEK`^lOD(r&|s5ZQettc
zLS`;#WLF_4KQSd9G(Q541#H?%Qj1H95t`FeOTblMacW6?L1IY;%pi1eoaVuu0b)SM
zykLfaCO;rEehBTEIjIV^3hLmT9A5!)t6o8N4kYJ;+zT>B!4}~W4Rn`iq-Ex$YC>Za
zqkRPKC2V7WPOUJeFiJ8sGZqz;Fl8~<Fr+coFwSG@V_;+`VW|Pjl`v$nHZv7DfoPBl
zh-fntBSYaPBsFZ!Ohus44K@rl>?us3hGL#m2}=!V3It@@B$#Os`)Zi_7#SJz6rd_V
zW^DpBo<U|Is{qfTvJly*tWpNgvqD-=pd<(y^+)eU`)P6%8G@P%Mj!&zJ}3gU9zf%;
znR&$}iFwJXx44pvi{f(=ONufpK<s2FyT}C8ZDWQso|qGh(r>Y4mnRmb7o#<Wzy@9e
zwK_mu69y(eMgc|vMiCHZ;$q@q6l3IJtdav4;o$i&NXG=+Zh*Hb{WMw88acPv<Kt8E
zljGyT#v@q?Zt{RzJN66=4D&$tf?CBKW-LraAQ6=I4yerxYMK>;T4^Zl9ZB%S9b*(r
zDs<|unGrO32bv9!VsB?iV+7BIb3}1~*`V2PjwnvZjCT}Q3V(_~3qurlieQRR3qur7
zig1cZ3qurdifD>h3qurNig=1d3qurtI|B<tlt3_prqnHdL;!%jo~X-C-?kJT+EKXM
zQ25$Uc-l^=?Izl^nQ%KAS6c~tI|;Roq^m%W_7TYK;B<wzZ3M0rp*1DA(MDpsh-z&j
zYP5%_(iWmpJBY|O5Gm~+w6+ht*`o_-6o8@<8k5Agd5CE5P}tTPZRgOmodas^f*M8z
zMVWae8ldKpUVJ=wh%i1L((|wZ^AyxU?LKuocvB19Rs^M)Vg*|Ta6tjCfx#6J*c^y%
zkl6^Odf>zWHbcYE$UsNI$ix7vI#8^Gm4R%H2k|sOMTKE3R@Gor!Odx~o@lVqhDHXl
zU@n>yph*@iWdv;yBZe_SW6_{-X2>umGqz#O6wn|hb1i!cOD#tUQw?JYb2C#>38+2S
z%vj4-VFen_Y-Vs_h}DZ>s^zTZssRmUvea<Zur@Q-veqzXvDa{AailPT+K!A2H7qGi
z<_xtgg;wF9flAJ1hFW&8sSH_+%}n_dYM4@(vpMo7famETGhte=0mfRk5{4}98nzTB
zanQgcQwn1ZQwpOELq$OiqYFc9MlBb3AeE<vtC_KuE1V%u1T>7w$WX!y8%WJ#ieSKS
z5!7`xOrSYDaRv|#vcU#4h09ih@CnT2yTTbJFy<K{Ts?uY=pEF(EFd=66?~w<-daYS
zt^oOya{^=045(`Guq=vdHc(qQk^$uAyZ}hJOkgZ}0S*l&Ne0l|=mf^1D<uqBJkSAP
zkSIHJ0GLUV0cijjNd;m6I4`f#p@bm|;!@Tc)-0|RP>3@XNz|}r@ux7=ux4?9+x$%6
z<w+njKq4TMVXoC-VJH#E=9<7%6vfC;%gxA8#aAMj!U1x5I71|8aTRkdcO7>G*cUYn
zSwb}sQ$eW#dB7QF^NJddEa4Q+X2uCjMJrOcQn*_fN<_ee&P*xH$b-%$qBX23Jk3lK
zn2OGnh-Hb_ux5c=E!oVJB?VeoRm+nX09M1>%rt?yhy%>#gR)sbK5J%7Va#R$%Ynjx
zHA|wI5v~_BXfB;4Q^SxVP{NQUo5J7AT+36#QNt52SHo1p6E9!GT*DKuP{UTk6R%jq
zUc(d5S;G^|peaaXj{`h_4H?UZ4fsMA8=-eQh#lpw5)4WO7YquO`K3kR;o~AL1_lO<
zP8~^u-bLD=&W;X<&;=3TE(4eV_Z##W7#N;0GB6Z_Mnl;cs<hEcM!ooW&~Px;f$m!@
zxrxQuMHV1qLF3YzOpxJe(5#9k6L_#6OlT^BN4$&lK?*@5uD4h~rDPFk^6VCCW=U#p
z@hzeFg8TyTQdY2ipmO~dTVg>$YF-N1&|55^G8Hl|&I%q!F9MH(gZgDf#-Pz__995=
z-QskuNKP#PEs2DThJyymi%dawgGQ!rv1cR}Czg~H-D1j3yu}0ZL23$Qtu4eS?jXZh
zKo!L;j-u3La3c}a0RsoaE!MQ0{KOK(0u1oTc@b#(PLuZ*b8=2G#1Ko6b)Z4@A}bKf
z8bp8;71@ATwje8*vr{W?ait`J*TIyg7T;n_y~UiLl!es21IJJ(0|Ub{Pz*JII&aW%
zbr#ULIwKb&2cr-psE^6R$ib+<$i>LV$ic`39cu@POMzvjU?NN~IRQp4Mgh<`JR=(u
z6C)QR52FyH1fvk66bEQjUV>4Eg@;jsk%v)+QG`hZIyw)YC=|z$KH$@8pk5}Z4+^f2
zR)Lq%fEUr&l`zyWmM~^9H8T}!f!WNUrJcoEC9GNCl_Et3HH@`Ph0AIfv)DmXZQ!Y|
z41~&}s1oKZ4$$0A8o2Dr^8w3&O)M%aVMLKLfSWp@gcU_b3}y<<7KmvzpfwasevtIa
zSp-U@MS-Ab;`U7}&Q47MHxrPOF=8A6lpAlcWfYerCTD9hL*hFU6wBa92j?^}0gg;B
z1_p*7pvVLz1_mY$M$nJ|8xtQh52FAh2y-zCFo~he$AWFv!^|Ei3zy&_9tH|=#uSEZ
zjv}5EhF}Iz7^N_#Ff@Z!i!g$DOi&&Zn8ys|F@t$5P#z1I#|q`Kf=tq6^Ft)|A|H?s
zK~qfNtf0wui#<KH1eC3Ct6>6<o)krcbb><!On`#rmQXw-2gavm<|XE2R;9+r^Dr?m
z6oaP07#KMiSeW=2ArN_{8X7VfT2Ts2P<jO6Vgv9fSP5u>F;f;(Gea#SWONI(xHpYa
zoS~Mfh7r`psbNfEmSlh~hGp>sPe6i`1-Qi5WVyv!Tw0J?ROALq9bED8MX4#J$)IU)
zj(E`UT}gbrCS;r_inky$IXfpcCBC>Mu_QH$6I6nLnV^|%c%v1P-$B7pl*quqAPVv~
zXfX@}6CaZRQ<W5EhJ)FNkpjT}1Qqq5rE6Ir|1m?_Y$Z%3%ry+nOrQzu6s8mwa6Qb(
z$N-L4R>-n4h#x`ALBVTGG+B$1Kt5zEEdZ_2D@p)G8`#f0@$sO6R*;|K<250%dW#k6
zu_Dlr6Oxxw85kI3L0$r-PzFW;#ww&GXRruF7zZy#K*~XDk-)_WsFDKB%7YhFF=R1=
z8cPr_G&921E{ijyFr>iO-D$$IG7C7*-eN8;%>`Ez;N+>v0!fPCtjnC7p96M&Q3=G|
zh-3*$)A7mqIf@7mNHJE4qsItnNH#e?2Ob?Dw}WC3>~<GuvddyBVFV3vX0af<9#qJc
zfEug}ps0~#05^-7YoLCqVX9$<tt5o|h!r$@T>|!2Q93A6K(!x+w~8`Bk;hV8l3#%C
zD-Mu*LG_y|$Xg(XGBC<93NcnmVDlJAAv{(<F#^&6_8Dl)O$h_2NXOwbP?5r1gX}Yw
z8kQPn2?p@`1V%}QY^I_-FdMvjKZ`Ak6_f!XqVO<au3-U%fek~RObL4ma}A3aXstQ}
zOtlz;1OuqC3Qo!_kOCZ>Q>#=Vjj8-Ja8pVlB|jCi@H;oLBsoJtLp4Q5K{Z8FlcOjb
z6pEZ60yR9r4G&0t4w_ciWJRPJh$twGi*i63!D$9efHLwe4$%AoDCBfNK@4gSL4sI=
zQH+TROF+YvV-ILhrGY#7qsxGLKJd)J0uFv~5Y|AKq-ruF1(GH!wuA)s9kO?CF&8D4
zgOe0i?}GD5QDV6%$hV;M!N4fRD2F4b6eX6!y@`@jO29b<w2BteurJD}VafteOVluA
zL0Q>MMFpS&8?*|ymbru_3$z9XwBUmo<ZaNh(q0xuP`8g6v>GQ1T%b2If<#ytve-c#
zS0;vdcJM9;@ZJz`s$oSA5XM_9i4~c}RZ@Pb<=|;L&_ujCXbVDSNoHbBVpVF2CMUS*
zij+LSEj><9j$_NrD^4vcDe?xT3Q#2vZi7M#oLh`#;Gm6S0}r}kuA@jyfh^*+0tF<b
z^5bI^W8?z2%><Y*b0|y$N|gr+Rd5ak^&T=AY8fjON<d-4*bM4q3xW52FoD)8)iBmD
z*)S9efI9NQ44~x|%(qy<yB5GL+M-;<q(>B}$(~;XDlyT^6ZZI$d{BGd1LSB>{lUP<
z!YBYLP%uhLs7jDcf@pIbMXsO_12uMw)IlszqA!XEu|TzA5vU$1$^vmgjw;Fnu?j#0
zDAg2|fr6eFTBRnZ=H$f3BV4J;2&t4nITKuKg8~b@=!3(Fg^??WM}&iuhl7WOgN1{g
zgOh`mgAbw#)ZB()P>_Py#h~^(%G?QLP6V{^i4%4HB$X|TJ%t^#5ehV$0-pWg2xida
zyv2nWA4{Cb1fBSBMBWbwn~j1F_KePc!lrR?&VG_TDhL`3M8pT$I3Q?eh6ZGI7&43p
z8(IVp5TeaHgSxEH312<5id5YW8k4#To}i^+&}nS&6adzBKp^jc%u}!e&xM1B{vh*>
z3RVjFB^jwj#R`c<>7}{gJ)U4`0)ypX(-ah}6sn<%B|ytTY8B82LcvD)!CNw*1|<lC
zmcAB)8kV5?9^8-wmmi=8WDP?+a}B7&!C9mM8Bzp^fXh`z&`c({-3VHT1)jUS#R{HY
z2WK#7GZ|bW7J-^cu#5_?N5NVO;l&Dz03#bC2V<2mBEsO4rl33x%AgFO6@nlPE&xGw
z5vVN;D%>D;G8R>W3N=PZRSvcZOn?gq&`3r#$U4wq2m?zMKaxeD0ECANNFHQ1*b>mF
z0;s!G!U&pH2h~X>Oj&GcEX_>CN+s-B9H3#uA}#Quea<|N60Q_RczcD(ugVCrd`AJY
zP!P02BN07Pz{?>^b73o_^fb94{ZmkD98!IQM^-@H)uI?sNywg_nwMIXnT#~s4e~W)
z5*yNBhlD!V>tF)z|3;AiL4IRkQexy_0u|MqSc_`#)Hw%ak6=k9*t}b8#i=DFsYONQ
zppqGEE{K4|E7%TDQV0au0jeoD<XD*axOmu#K%$z$;5y(Ib7E3*5h#RjvF3q%4ff(K
z4lr9UHxbe;=P6FkEU46jO!(hogU}HFf)XrvwO^4V$Qe!`0yG&88Gix?8fY5~*iJ+(
t0bzlhc#Fd(7rbBF4zw_^7&MX3!U!5*;bP=s5{lweld5NM6RBs&2LM(YYy<!R
delta 4358
zcmbPMdp5Y<n3tDprMGaLh#>>RV+JI^%D}+jz`(#z3?fn(q8L&bQ<!oXqnN-na~4Y!
zi#tOKa|%lfLkdeOOBQP?Qx;n@OBB01LkepOTMI)9TPjC0a};L^Q!s-jd*WIq28NJ~
z%wmPy{FKt1RE6aHypqJsjb|4yY5FB2*~`ET<rFtDFfgPtL@}l?fGmw-PLZCRATJ`-
z&XC5K!k)s>!V$%u!kNO=!Vtxg!kxm?!Z3M~yp~)lYl?V^L<>VRV-$aiWQtS^LzF-}
z0}DfxU@(KG%;t0QzU=lq3=9k)><sds8xsRVIzugU3S%ux2}2fRIzuyKEo%u=4P!H7
zEn5mxEqe-6Ek_A+4PyyQGgDDT3De{-b^Ur~Nrq-dMur-e6lQaVT9!h)aE2PjEY4<z
zT6VBnhAhTrrlM;#Oesv+97We^7_+!in6p75j9DBhjFJqsY!yZ|j4lkZTD4p?%vtO;
zTv;3`FjE<_*lJj6xN2CNnQK|$b}S2L$a?@Wy&hzI{wt`lEE5=uwt=nUoWNM5Q^{At
zki}iY)XZ4R6wZ(*!pH!EB|IrC&5TWqkqmiE5ezj<5ND(?OEA>3<)wjjb4*|?np6XF
zAhRSx3UfBd+>8>2EZ!7GxG1X>$fNaOx27|LtjH5VRgw<2xP~E%uZ9Vxt(hr?sg|jh
zErl(Gy@jEKzlN!XDTO(Osg+5Rp_$1A6z0q|jJ2#aps*IGVNKy^W}3iQbgM)#OQ?o5
zOSpzLOQe}8OSA^0k~=S;z67L*vzciEQ;`^$%>`vM)v#u9G&80!X0w3hKq0}JCEUyi
z*9>y5SeAGVLmEQ~LzYAecQ11-cMV4kcf4c`a}9UAR1I4Vcf52BdkuFyXAO5SgC<Xv
zf^&XeaY<2Wa!G!XLVlXU<XSB;HoupQ3=9mD_i8CKy=0pFOUsmHB|{PGWJeJZHc<8{
zVx64JnZWmp%O)qYxFk70x4^EMVRC}DjpHqrl*E$6Tdc(yi3O>eOhvp53=FqeQ%Wie
zQZ;36G36E9V$VxSEGkN@yv3Z8ns<vUvp6%axFj(zIrSDxN@jA&<XhS*T3}r)*{PMq
zw^;I0GIMXSCl*6AafVbDq`DRr<rm#z1&K^P!zsghi@7SZV6wN43>zrz7jaL{)lsg$
z#aWh_T3#HVnU`{l3+x;ayW|#oUOZSrljjz5a!zp(NXsn_kPt{|ks8>+nRzLx6}LF@
zQp@8({wY!iiL#euq=MbV3sW6moSKt%i#a>B@)lQ0Vo7{iVoqsl@h#^3q^$ZPNI)?#
zFcfiu2xSHah6GTyi~$t`Oe~Bnj9iQyj4)b=k&jV;k%y6kQI3&|k&97)k&ls&QG$_+
zk%^Iykqs;h5>;XnVPs<zV3YyrWZ_|yVB}$xVH5$2vN3{SmAG$dPRZncUCsJB1_p*2
zh7`tZrXrgXh8o5a#w@00reXy!n>h=V>(y!)YnckGY8bOv!Lp3COeKt2Y&A^r?0I%z
zbr3xvU^YZgN(o~Y2Pg-mf%8M27>o}y10r5i4=Ny;{E9$nNt3fk2owzRppfMDO)Sn%
zO>qRJgCc&AumFfq1j(`G<>w|AXBTmUIT^(ziOJcT%tcxt1^OUD8$_T5t|$Wo!!%If
zg6v>m;$W0uWMkrCVdP?xnEYDLsvcDI7Ds?WgE566o1=&)g&~-sgrS77hM}1$3tS>H
zGJwjT6vh;WX671(cor~^3Cd#y^O&JLHZYF`%3}warOE0CPCDR}QX~#?qACLegC@r<
z*5cBF`qZK#CXj!G;^RT7wlY2~GcPeGvnn+{{uX<BYDr>ANfB24Ot)A-X{^WqWEkTu
z&iMG`oW$bd_;^V20;eof1_p+uAkXuGQWhfz0}C|u@G({?fzm?eWI+QtF;EzQ6tFNb
zFo1#&Tt^s8GLmP?Vwo&&B<;frj*45%#ihALk{};|@&Y)2XtETUg9N}4#GIU;bBndO
zB(bOj;s}so;4}{QN_=vD&RT}ajYd|CZj<*InKEUuPX1^l9msZzIWw;WY(<eJ$S~Yi
z6j_79ilw+DzX06^j`Y+LaQ@xMFge0lBM=m-AWwm4HU<U;u%~=L5zUYVN%%@N3|WjR
zOrQd}h5;1Sk_>4~k_^ENnBpM$$t$@c>u<4V=A~pNrxyDafkI7_`4$&A%jf2&q{f3o
zqzD|lOyFR-#gUnp0(KO-YkA{SD@uwIlOayt28tt)BN>>u7$umhWWjOco1c=H1I`J0
zFx8W*jHMYTOx|OpFnOD?iflm*Qx<CqlO)K+Y!Fs9Q&B+;Qw>8F+vKmt){M-P4NY|G
zZ9$<5&P{0H3o1^*uD-<;1}dSz<)<bq*zrY*AYGtB5FF}7b|5Y|M!^Iqf!*RLh6egx
zkPkp=85rdlg%~v$C775Pt0W-)n0(EoJ_H<Gkc5OamXR#I#axtFj@wdDwumoEEI-CD
zd8O%IM&`-sW?J=vAlJdY1IaNUAKzlj%qvbUDk%~JDNqLyE}$F(Nw>Ec%OKuogG4KO
zG;t)RK;rK-$bXPr#K$Pc$i*nZD8wkh1oL73WC8OeQ?TjwAk&$_AqcLw@*&{}4l^(T
z4mkGsl6+95c9UW9a&wLPWKb#vrB+ZVfmDH0YZn6psI~?bMU1r|7ARhsOBle#b5#vP
z7Av>_uPOmGofuM>QaC|P3sA8RO4E`I*{scsMP4QBDa@cYPz^&0izGt|OEyDsQVl~2
zmn1_CV>U}sRtbBGIyXc?76(`fqXa`OQyxbNdkPOkyoM<aRABQ?&SiCADq^0T&09YC
zIj=6)%m4rX|1SyvH3%9x+$S$%*RI#(gk)rCP)6m-EG{lh0@v|Hpuhv=o|4q`)S@B@
zkTi2*US$y|Ho%PmO=fV`xy6+ipO%_fQd*Q+3@I}~;Q@&lP|y|0FfcIO2So=cMj4p7
z7^N8H7(wMa6C)d=xBwFqBO6ndCM4y-bI@c*E@^l<2g+$6L&1UEHo2EuroIVO&4EL+
z2^^ZBazO+fnv9^vKZu1EnvCF}f`lexil`(*4HGmp#Xy0U!d}A=&jD&4gVlk8lz9R}
z-UpaH6Bvv5Y8bOXvB{Oq#8<qiggr$Zq_JXO3HM|t9{u`zCG05@Adwoz8Wu^0ES_d?
z<V|2KW`K(DHiIl>Em~8;o+1TOS-2*gAuj}M1%EcEEg4k;F4(}<uz*^Z#U>@}DFPt1
zHH=6m2GlTSK~0;$Sd>)4o+7OUQUp!}LLgR(FtUGYnQItp7~=VAKouhYWGP;6Mo6mA
zW?RWx1gb|SUu4&qyn!!}57d_eClN3)xsk(t^D=g2#(DunIw=YSWlK;3DGCCyK#8L$
z7{mf64o46Rl(lcM<tJsOCYRjeOwB9JO)W|+Nxj9Ko0)fuIXAHaoOr+qMw1Pk`yc`#
zAkCnR394jLZZVf7=77@@5152hPQsH51%(AZg3=VoLQo+r#V9vfS<p^E89f~p2#T<3
zGWux>O>P%d^=FTdPsvY?k1qnX;coF1mn4>CCg-M>WaOt5fm$9#pc=Ty2IK}%xlja3
zaYe2mt~-bTRYyhsAQmW57sX6gv=FujrA)BnG#QI5Kyu(z4=NyJK`xYHU|`^IVqxSm
z<KgGw;9%ik<zVOF<Y49C<6z?vDq@;kXpzA<ck(R@Ti(=?qRix+)Wo8^%)Io;3YL0k
zB95x=Xd)9;H*b!zN@Vml0a*;{qu%0(kI&4@EQyc5#R_Thf?T2rDd3=GXLfmFQF<}V
zxnTEmg7l{_Ffdp$PX1@@W`0Z17vUO6OGlHjC>2yzGD2bttPM<n^@8dWXU567HU^A3
zljqq;@)Rk6lN@^<M+wK|vo=ok&}JjZIgq3TZYhFVkwu1}(uF-eH7~U&GZ}145h&O+
zA+<8t)!-0@5ODiE85tOgL4^YYlLR9N6DXQFi()}Oi31VwAOh4-F9Lb2{uW0;QGP*c
zQAs7(z*}s^sU;<;MMX&<Wnhnh2w13qEdd3nB*+pDP$bH+F!6EmuoZzsHMxpFCf{Ps
zE6pva1pDU}2biswn^*+*R1w&ZU@uz<gX{t6h4=zw8H5G0>K2DhE_l$v4wRXSxfmE2
OSQt4NxtPW38EOF|Y|w@P
diff --git a/multimodal/datasets/base.py b/multimodal/datasets/base.py
index 0e033e7..3d58c39 100644
--- a/multimodal/datasets/base.py
+++ b/multimodal/datasets/base.py
@@ -1,7 +1,7 @@
from __future__ import print_function
import numpy as np
import numpy.ma as ma
-from metriclearning.datasets.data_sample import DataSample
+from multimodal.datasets.data_sample import DataSample
from six.moves import cPickle as pickle #for performance
diff --git a/multimodal/datasets/data_sample.py b/multimodal/datasets/data_sample.py
index fbcf5aa..ed3ab6d 100644
--- a/multimodal/datasets/data_sample.py
+++ b/multimodal/datasets/data_sample.py
@@ -22,11 +22,337 @@ xxxxxxxx xxxx xxxx xxxx
the number nbL and nbEx and , the fourth dictionaries for sample,
prefix, suffix and factor where they are computed
"""
+from abc import ABCMeta
import numpy as np
import numpy.ma as ma
+import scipy.sparse as sp
+class MultiModalData(metaclass=ABCMeta):
-class MultiModalArray(ma.MaskedArray, np.ndarray):
+ @staticmethod
+ def _first_validate_views_ind(views_ind, n_features):
+ """Ensure proper format for views_ind and return number of views."""
+ views_ind = np.array(views_ind)
+
+ if np.issubdtype(views_ind.dtype, np.integer) and views_ind.ndim == 1:
+ if len(views_ind) > 2 and np.any(views_ind[:-1] >= views_ind[1:]):
+ raise ValueError("Values in views_ind must be sorted.")
+ if views_ind[0] < 0 or views_ind[-1] > n_features:
+ raise ValueError("Values in views_ind are not in a correct "
+ + "range for the provided data.")
+ view_mode_ = "slices"
+ n_views = views_ind.shape[0]-1
+ else:
+ if views_ind.ndim == 1:
+ if not views_ind.dtype == np.object:
+ raise ValueError("The format of views_ind is not "
+ + "supported.")
+ for ind, val in enumerate(views_ind):
+ views_ind[ind] = np.array(val)
+ if not np.issubdtype(views_ind[ind].dtype, np.integer):
+ raise ValueError("Values in views_ind must be "
+ + "integers.")
+ if views_ind[ind].min() < 0 \
+ or views_ind[ind].max() >= n_features:
+ raise ValueError("Values in views_ind are not in a "
+ + "correct range for the provided "
+ + "data.")
+ elif views_ind.ndim == 2:
+ if not np.issubdtype(views_ind.dtype, np.integer):
+ raise ValueError("Values in views_ind must be integers.")
+ if views_ind.min() < 0 or views_ind.max() >= n_features:
+ raise ValueError("Values in views_ind are not in a "
+ + "correct range for the provided data.")
+ else:
+ raise ValueError("The format of views_ind is not supported.")
+ view_mode_ = "indices"
+ n_views = views_ind.shape[0]
+ return (views_ind, n_views, view_mode_)
+
+ def _extract_view(self, ind_view):
+ """Extract the view for the given index ind_view from the dataset X."""
+ if self.view_mode_ == "indices":
+ return self[:, self.views_ind[ind_view]]
+ else:
+ return self[:, self.views_ind[ind_view]:self.views_ind[ind_view+1]]
+
+ def _validate_views_ind(self, views_ind, n_features):
+ """Ensure proper format for views_ind and return number of views."""
+ views_ind = np.array(views_ind)
+
+ if np.issubdtype(views_ind.dtype, np.integer) and views_ind.ndim == 1:
+ if len(views_ind) > 2 and np.any(views_ind[:-1] >= views_ind[1:]):
+ raise ValueError("Values in views_ind must be sorted.")
+ if views_ind[0] < 0 or views_ind[-1] > n_features:
+ raise ValueError("Values in views_ind are not in a correct "
+ + "range for the provided data.")
+ self.view_mode_ = "slices"
+ n_views = views_ind.shape[0]-1
+ else:
+ if views_ind.ndim == 1:
+ if not views_ind.dtype == np.object:
+ raise ValueError("The format of views_ind is not "
+ + "supported.")
+ for ind, val in enumerate(views_ind):
+ views_ind[ind] = np.array(val)
+ if not np.issubdtype(views_ind[ind].dtype, np.integer):
+ raise ValueError("Values in views_ind must be "
+ + "integers.")
+ if views_ind[ind].min() < 0 \
+ or views_ind[ind].max() >= n_features:
+ raise ValueError("Values in views_ind are not in a "
+ + "correct range for the provided "
+ + "data.")
+ elif views_ind.ndim == 2:
+ if not np.issubdtype(views_ind.dtype, np.integer):
+ raise ValueError("Values in views_ind must be integers.")
+ if views_ind.min() < 0 or views_ind.max() >= n_features:
+ raise ValueError("Values in views_ind are not in a "
+ + "correct range for the provided data.")
+ else:
+ raise ValueError("The format of views_ind is not supported.")
+ self.view_mode_ = "indices"
+ n_views = views_ind.shape[0]
+ self.views_ind = views_ind
+ self.n_views = n_views
+ return (views_ind, n_views)
+
+class MultiModalSparseInfo():
+
+ def __init__(self, data, view_ind=None):
+ """Constructor of Metriclearn_array"""
+ shapes_int = []
+ index = 0
+ new_data = np.ndarray([])
+ n_views = data.size
+ thekeys = None
+ # view_ind_self = None
+ view_mode = 'slices'
+
+ if (sp.issparse(data)) and data.ndim > 1:
+ if view_ind is not None:
+ try:
+ view_ind = np.asarray(view_ind)
+ except :
+ raise TypeError("n_views should be list or nparray")
+ elif view_ind is None:
+ if data.shape[1] > 1:
+ view_ind = np.array([0, data.shape[1]//2, data.shape[1]])
+ else:
+ view_ind = np.array([0, data.shape[1]])
+
+ new_data = data
+ # view_ind_self = view_ind
+ view_ind, n_views, view_mode = self._first_validate_views_ind(view_ind,
+ data.shape[1])
+ if view_ind.ndim == 1 and view_mode.startswith("slicing"):
+ shapes_int = [in2 - in1 for in1, in2 in zip(view_ind, view_ind[1:])]
+
+ if data.shape[0] < 1 or data.shape[1] < 1:
+ raise ValueError("input data shouldbe not empty")
+ self.view_mode_ = view_mode
+ self.views_ind = view_ind
+ self.shapes_int = shapes_int
+ self.n_views = n_views
+
+class MultiModalSparseArray(sp.csr_matrix, sp.csc_matrix, MultiModalSparseInfo, MultiModalData):
+ """
+ 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
+ ----------
+
+ view_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))
+
+ """
+
+ def __init__(self, *arg, **kwargs ):
+ """Constructor of Metriclearn_array"""
+ if sp.issparse(arg[0]):
+ MultiModalSparseInfo.__init__(self, *arg)
+ if isinstance(arg[0], sp.csr_matrix) :
+ sp.csr_matrix.__init__(self, arg[0])
+ elif isinstance(arg[0], sp.csc_matrix):
+ sp.csc_matrix.__init__(self, arg[0])
+ else:
+ raise TypeError("This sparse format is not supported")
+ else:
+ if isinstance(self,sp.csr_matrix):
+ sp.csr_matrix.__init__(self, *arg, **kwargs)
+ elif isinstance(self, sp.csc_matrix):
+ sp.csc_matrix.__init__(self, *arg, **kwargs)
+
+
+# class MultiModalSparseArray(sp.csr_matrix, sp.csc_matrix, MultiModalData):
+# """
+# 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
+# ----------
+#
+# view_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))
+# >>> print(data.__class__)
+# <class 'dict'>
+# >>> multiviews = MultiModalArray(data)
+# >>> multiviews.shape
+# (120, 240)
+# >>> multiviews.keys
+# dict_keys([0, 1])
+# >>> multiviews.shapes_int
+# [120, 120]
+# >>> multiviews.n_views
+# 2
+#
+#
+# """
+#
+# def __init__(self, data, view_ind=None, shape=None, dtype=None, copy=False):
+# """Constructor of Metriclearn_array"""
+# shapes_int = []
+# index = 0
+# new_data = np.ndarray([])
+# n_views = 1
+# thekeys = None
+# # view_ind_self = None
+# view_mode = 'slices'
+# if isinstance(data, tuple) and len(data) == 3:
+# data_data = data[0]
+# indices = data[1]
+# indptr = data[2]
+# data_shape = shape
+# else:
+# if shape is None:
+# data_shape = data.shape
+# if dtype is None:
+# dtype = data.dtype
+# data_data = data.data
+# data_indices = data.indices
+# data_indptr = data.indptr
+# if (sp.issparse(data)) and data.ndim > 1:
+# if view_ind is not None:
+# try:
+# view_ind = np.asarray(view_ind)
+# except :
+# raise TypeError("n_views should be list or nparray")
+# elif view_ind is None:
+# if data.shape[1] > 1:
+# view_ind = np.array([0, data.shape[1]//2, data.shape[1]])
+# else:
+# view_ind = np.array([0, data.shape[1]])
+#
+# new_data = data
+# # view_ind_self = view_ind
+# view_ind, n_views, view_mode = self._first_validate_views_ind(view_ind,
+# data_shape[1])
+# if view_ind.ndim == 1 and view_mode.startswith("slicing"):
+# shapes_int = [in2 - in1 for in1, in2 in zip(view_ind, view_ind[1:])]
+# if isinstance(data, sp.csr_matrix) :
+# sp.csr_matrix.__init__(self, (data_data, data_indices, data_indptr), shape=data_shape)
+# #sp.csr_matrix.__init__(self, data)
+# elif isinstance(data, sp.csc_matrix):
+# sp.csc_matrix.__init__(self, (data_data, data_indices, data_indptr), shape=data_shape)
+# #sp.csc_matrix.__init__(self, data)
+# else:
+# raise TypeError("This sparse format is not supported")
+# if self.shape[0] < 1 or self.shape[1] < 1:
+# raise ValueError("input data shouldbe not empty")
+# self.view_mode_ = view_mode
+# self.views_ind = view_ind
+# self.shapes_int = shapes_int
+# self.n_views = n_views
+
+
+class MultiModalArray(np.ndarray, MultiModalData):
"""
MultiModalArray inherit from numpy ndarray
@@ -98,9 +424,10 @@ class MultiModalArray(ma.MaskedArray, np.ndarray):
shapes_int = []
index = 0
new_data = np.ndarray([])
- n_views = len(data)
+ n_views = 1
thekeys = None
- view_ind_self = None
+ # view_ind_self = None
+ view_mode = 'slices'
if isinstance(data, dict):
n_views = len(data)
for key, dat_values in data.items():
@@ -110,38 +437,65 @@ class MultiModalArray(ma.MaskedArray, np.ndarray):
thekeys = data.keys()
if isinstance(data, np.ndarray) and view_ind is None and data.ndim == 1:
n_views = data.shape[0]
+ view_ind = np.empty(n_views+1)
+ view_ind[0] = 0
for dat_values in data:
+ try:
+ dat_values = np.array(dat_values)
+ except:
+ raise TypeError("input format is not supported")
shapes_int.append(dat_values.shape[1])
+ view_ind[index+1] = dat_values.shape[1] + view_ind[index]
new_data = cls._populate_new_data(index, dat_values, new_data)
index += 1
- elif isinstance(data, np.ndarray) and data.ndim > 1:
+ elif (isinstance(data, np.ndarray) ) and data.ndim > 1:
+ try:
+ data = np.asarray(data)
+ except:
+ raise TypeError("input format is not supported")
+
if view_ind is not None:
try:
view_ind = np.asarray(view_ind)
except :
raise TypeError("n_views should be list or nparray")
- n_views = view_ind.shape[0] - 1
elif view_ind is None:
if data.shape[1] > 1:
view_ind = np.array([0, data.shape[1]//2, data.shape[1]])
else:
view_ind = np.array([0, data.shape[1]])
- view_ind, n_views = cls._first_validate_views_ind(view_ind,
- data.shape[1])
- shapes_int = [ in2-in1 for in1, in2 in zip(view_ind, view_ind[1: ])]
new_data = data
- view_ind_self = view_ind
-
+ else:
+ try:
+ new_data = np.asarray(data)
+ if new_data.ndim == 1:
+ new_data = new_data.reshape(1, new_data.shape[0])
+ view_ind = np.array([0, new_data.shape[1]])
+ except Exception as e:
+ raise ValueError('Reshape your data')
+
+ # view_ind_self = view_ind
+ # if new_data.shape[1] < 1:
+ # msg = ("%d feature\(s\) \\(shape=\%s\) while a minimum of \\d* "
+ # "is required.") % (new_data.shape[1], str(new_data.shape))
+ # # "%d feature\(s\) \(shape=\(%d, %d\)\) while a minimum of \d* is required." % (new_data.shape[1], new_data.shape[0], new_data.shape[1])
+ # raise ValueError(msg)
+ view_ind, n_views, view_mode = cls._first_validate_views_ind(view_ind,
+ new_data.shape[1])
+ if view_ind.ndim == 1 and view_mode.startswith("slicing"):
+ shapes_int = [in2 - in1 for in1, in2 in zip(view_ind, view_ind[1:])]
# obj = ma.MaskedArray.__new(new_data) # new_data.view() a.MaskedArray(new_data, mask=new_data.mask).view(cls)
# bj = super(Metriclearn_array, cls).__new__(cls, new_data.data, new_data.mask)
+
if hasattr(new_data, "mask"):
obj = ma.masked_array(new_data.data, new_data.mask).view(cls)
elif hasattr(new_data, "data") and \
hasattr(new_data, "shape") and len(new_data.shape) > 0:
- obj = np.asarray(new_data.data).view(cls)
+ obj = np.asarray(new_data.data).view(cls)
else:
- obj = np.recarray.__new__(cls, shape=(), dtype=np.float)
- obj.views_ind = view_ind_self
+ obj = np.recarray.__new__(cls, shape=(0, 0), dtype=np.float)
+ obj.view_mode_ = view_mode
+ obj.views_ind = view_ind
obj.shapes_int = shapes_int
obj.n_views = n_views
obj.keys = thekeys
@@ -150,47 +504,60 @@ class MultiModalArray(ma.MaskedArray, np.ndarray):
@staticmethod
def _populate_new_data(index, dat_values, new_data):
if index == 0:
- if isinstance(dat_values, ma.MaskedArray) or isinstance(dat_values, np.ndarray):
+ if isinstance(dat_values, ma.MaskedArray) or \
+ isinstance(dat_values, np.ndarray) or sp.issparse(dat_values):
new_data = dat_values
else:
- new_data = dat_values.view(ma.MaskedArray) # ma.masked_array(dat_values, mask=ma.nomask) dat_values.view(ma.MaskedArray) #(
- new_data.mask = ma.nomask
+ new_data = dat_values.view(np.ndarray) # ma.masked_array(dat_values, mask=ma.nomask) dat_values.view(ma.MaskedArray) #(
+ # new_data.mask = ma.nomask
else:
- if isinstance(dat_values, ma.MaskedArray) or isinstance(dat_values, np.ndarray):
+ if isinstance(dat_values, np.ndarray):
+ new_data = np.hstack((new_data, dat_values))
+ elif isinstance(dat_values, ma.MaskedArray):
new_data = ma.hstack((new_data, dat_values))
+ elif sp.issparse(dat_values):
+ new_data = sp.hstack((new_data, dat_values))
else:
- new_data = ma.hstack((new_data, dat_values.view(ma.MaskedArray) ) ) # ma.masked_array(dat_values, mask=ma.nomask
+ new_data = np.hstack((new_data, dat_values.view(np.ndarray) ) ) # ma.masked_array(dat_values, mask=ma.nomask
return new_data
def __array_finalize__(self, obj):
if obj is None: return
- super(MultiModalArray, self).__array_finalize__(obj)
+ # super(MultiModalArray, self).__array_finalize__(obj)
self.shapes_int = getattr(obj, 'shapes_int', None)
self.n_views = getattr(obj, 'n_views', None)
self.keys = getattr(obj, 'keys', None)
self.views_ind = getattr(obj, 'views_ind', None)
+ self.view_mode_ = getattr(obj, 'view_mode_', None)
+
+ def __reduce__(self):
+ # Get the parent's __reduce__ tuple
+ pickled_state = super(MultiModalArray, self).__reduce__()
+ # Create our own tuple to pass to __setstate__
+ new_state = pickled_state[2] + (self.__dict__,)
+ # Return a tuple that replaces the parent's __setstate__ tuple with our own
+ return (pickled_state[0], pickled_state[1], new_state)
+
+ def __setstate__(self, state):
+ self.__dict__.update(state[-1])
+ super(MultiModalArray, self).__setstate__(state[0:-1])
def get_col(self, view, col):
start = np.sum(np.asarray(self.shapes_int[0: view]))
- return self.data[start+col, :]
+ return self[start+col, :]
def get_view(self, view):
start = int(np.sum(np.asarray(self.shapes_int[0: view])))
stop = int(start + self.shapes_int[view])
- return self.data[:, start:stop]
+ return self[:, start:stop]
+
- def _extract_view(self, ind_view):
- """Extract the view for the given index ind_view from the dataset X."""
- if self.view_mode_ == "indices":
- return self.data[:, self.views_ind[ind_view]]
- else:
- return self.data[:, self.views_ind[ind_view]:self.views_ind[ind_view+1]]
def set_view(self, view, data):
start = int(np.sum(np.asarray(self.shapes_int[0: view])))
stop = int(start + self.shapes_int[view])
if stop-start == data.shape[0] and data.shape[1]== self.data.shape[1]:
- self.data[:, start:stop] = data
+ self[:, start:stop] = data
else:
raise ValueError(
"shape of data does not match (%d, %d)" %stop-start %self.data.shape[1])
@@ -220,63 +587,8 @@ class MultiModalArray(ma.MaskedArray, np.ndarray):
dico[view] = self.get_view(view)
return dico
- @staticmethod
- def _first_validate_views_ind(views_ind, n_features):
- """Ensure proper format for views_ind and return number of views."""
- views_ind = np.array(views_ind)
- if np.issubdtype(views_ind.dtype, np.integer) and views_ind.ndim == 1:
- if np.any(views_ind[:-1] >= views_ind[1:]):
- raise ValueError("Values in views_ind must be sorted.")
- if views_ind[0] < 0 or views_ind[-1] > n_features:
- raise ValueError("Values in views_ind are not in a correct "
- + "range for the provided data.")
- n_views = views_ind.shape[0]-1
- else:
- raise ValueError("The format of views_ind is not "
- + "supported.")
-
- return (views_ind, n_views)
- def _validate_views_ind(self, views_ind, n_features):
- """Ensure proper format for views_ind and return number of views."""
- views_ind = np.array(views_ind)
- if np.issubdtype(views_ind.dtype, np.integer) and views_ind.ndim == 1:
- if np.any(views_ind[:-1] >= views_ind[1:]):
- raise ValueError("Values in views_ind must be sorted.")
- if views_ind[0] < 0 or views_ind[-1] > n_features:
- raise ValueError("Values in views_ind are not in a correct "
- + "range for the provided data.")
- self.view_mode_ = "slices"
- n_views = views_ind.shape[0]-1
- else:
- if views_ind.ndim == 1:
- if not views_ind.dtype == np.object:
- raise ValueError("The format of views_ind is not "
- + "supported.")
- for ind, val in enumerate(views_ind):
- views_ind[ind] = np.array(val)
- if not np.issubdtype(views_ind[ind].dtype, np.integer):
- raise ValueError("Values in views_ind must be "
- + "integers.")
- if views_ind[ind].min() < 0 \
- or views_ind[ind].max() >= n_features:
- raise ValueError("Values in views_ind are not in a "
- + "correct range for the provided "
- + "data.")
- elif views_ind.ndim == 2:
- if not np.issubdtype(views_ind.dtype, np.integer):
- raise ValueError("Values in views_ind must be integers.")
- if views_ind.min() < 0 or views_ind.max() >= n_features:
- raise ValueError("Values in views_ind are not in a "
- + "correct range for the provided data.")
- else:
- raise ValueError("The format of views_ind is not supported.")
- self.view_mode_ = "indices"
- n_views = views_ind.shape[0]
- self.views_ind = views_ind
- self.n_views = n_views
- return (views_ind, n_views)
class DataSample(dict):
"""
@@ -285,9 +597,9 @@ class DataSample(dict):
:Example:
- >>> from metriclearning.datasets.base import load_dict
- >>> from metriclearning.tests.datasets.get_dataset_path import get_dataset_path
- >>> from metriclearning.datasets.data_sample import DataSample
+ >>> 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__)
@@ -315,20 +627,20 @@ class DataSample(dict):
# The dictionary that contains the sample
super(DataSample, self).__init__(kwargs)
- self._data = None # Metriclearn_array(np.zeros((0,0)))
+ self._data = None # Metriclearn_arrayMultiModalArray(np.zeros((0,0)))
if data is not None:
self._data = MultiModalArray(data)
@property
def data(self):
- """Metriclearn_array"""
+ """MultiModalArray"""
return self._data
@data.setter
def data(self, data):
- if isinstance(data, (MultiModalArray, np.ndarray, ma.MaskedArray, np.generic)):
+ if isinstance(data, (MultiModalArray, np.ndarray, ma.MaskedArray, np.generic)) or sp.issparse(data):
self._data = data
else:
raise TypeError("sample should be a MultiModalArray or numpy array.")
diff --git a/multimodal/kernels/__init__.py b/multimodal/kernels/__init__.py
index 8647836..7d48045 100644
--- a/multimodal/kernels/__init__.py
+++ b/multimodal/kernels/__init__.py
@@ -1 +1 @@
-__all__ = ['MVML']
+__all__ = ['MVML', 'MKernel', 'MVML']
diff --git a/multimodal/kernels/lpMKL.py b/multimodal/kernels/lpMKL.py
index bdcfe13..1eca272 100644
--- a/multimodal/kernels/lpMKL.py
+++ b/multimodal/kernels/lpMKL.py
@@ -5,7 +5,7 @@ from sklearn.utils.multiclass import unique_labels
from sklearn.utils.validation import check_X_y
from sklearn.utils.validation import check_array
from sklearn.utils.validation import check_is_fitted
-from metriclearning.mkernel import MKernel
+from multimodal.kernels.mkernel import MKernel
class MKL(BaseEstimator, ClassifierMixin, MKernel):
diff --git a/multimodal/kernels/mvml.py b/multimodal/kernels/mvml.py
index d42e9a4..535d974 100644
--- a/multimodal/kernels/mvml.py
+++ b/multimodal/kernels/mvml.py
@@ -10,8 +10,8 @@ from sklearn.utils.validation import check_X_y
from sklearn.utils.validation import check_array
from sklearn.metrics.pairwise import check_pairwise_arrays
from sklearn.utils.validation import check_is_fitted
-from metriclearning.datasets.data_sample import DataSample, Metriclearn_array
-from metriclearning.mkernel import MKernel
+from multimodal.datasets.data_sample import DataSample, MultiModalArray
+from multimodal.kernels.mkernel import MKernel
"""
Copyright (C) 2018 Riikka Huusari
@@ -196,7 +196,9 @@ class MVML(MKernel, BaseEstimator, ClassifierMixin):
# Return the classifier
self.learn_mvml(learn_A=self.learn_A, learn_w=self.learn_w, n_loops=self.n_loops)
if self.warning_message:
- print("warning appears during fit process", self.warning_message)
+ import logging
+ logging.warning("warning appears during fit process" + str(self.warning_message))
+ # print("warning appears during fit process", self.warning_message)
return self
def learn_mvml(self, learn_A=1, learn_w=0, n_loops=6):
diff --git a/multimodal/tests/test.py b/multimodal/tests/test.py
new file mode 100644
index 0000000..9a68d84
--- /dev/null
+++ b/multimodal/tests/test.py
@@ -0,0 +1,224 @@
+
+from abc import ABCMeta
+import numpy as np
+import numpy.ma as ma
+import scipy.sparse as sp
+
+from multimodal.boosting.mumbo import MumboClassifier
+
+class MultiModalData(metaclass=ABCMeta):
+
+ @staticmethod
+ def _first_validate_views_ind(views_ind, n_features):
+ """Ensure proper format for views_ind and return number of views."""
+ views_ind = np.array(views_ind)
+ if np.issubdtype(views_ind.dtype, np.integer) and views_ind.ndim == 1:
+ if np.any(views_ind[:-1] >= views_ind[1:]):
+ raise ValueError("Values in views_ind must be sorted.")
+ if views_ind[0] < 0 or views_ind[-1] > n_features:
+ raise ValueError("Values in views_ind are not in a correct "
+ + "range for the provided data.")
+ view_mode_ = "slices"
+ n_views = views_ind.shape[0]-1
+ else:
+ if views_ind.ndim == 1:
+ if not views_ind.dtype == np.object:
+ raise ValueError("The format of views_ind is not "
+ + "supported.")
+ for ind, val in enumerate(views_ind):
+ views_ind[ind] = np.array(val)
+ if not np.issubdtype(views_ind[ind].dtype, np.integer):
+ raise ValueError("Values in views_ind must be "
+ + "integers.")
+ if views_ind[ind].min() < 0 \
+ or views_ind[ind].max() >= n_features:
+ raise ValueError("Values in views_ind are not in a "
+ + "correct range for the provided "
+ + "data.")
+ elif views_ind.ndim == 2:
+ if not np.issubdtype(views_ind.dtype, np.integer):
+ raise ValueError("Values in views_ind must be integers.")
+ if views_ind.min() < 0 or views_ind.max() >= n_features:
+ raise ValueError("Values in views_ind are not in a "
+ + "correct range for the provided data.")
+ else:
+ raise ValueError("The format of views_ind is not supported.")
+ view_mode_ = "indices"
+ n_views = views_ind.shape[0]
+ return (views_ind, n_views, view_mode_)
+
+ def _extract_view(self, ind_view):
+ """Extract the view for the given index ind_view from the dataset X."""
+ if self.view_mode_ == "indices":
+ return self[:, self.views_ind[ind_view]]
+ else:
+ return self[:, self.views_ind[ind_view]:self.views_ind[ind_view+1]]
+
+ def _validate_views_ind(self, views_ind, n_features):
+ """Ensure proper format for views_ind and return number of views."""
+ views_ind = np.array(views_ind)
+ if np.issubdtype(views_ind.dtype, np.integer) and views_ind.ndim == 1:
+ if np.any(views_ind[:-1] >= views_ind[1:]):
+ raise ValueError("Values in views_ind must be sorted.")
+ if views_ind[0] < 0 or views_ind[-1] > n_features:
+ raise ValueError("Values in views_ind are not in a correct "
+ + "range for the provided data.")
+ self.view_mode_ = "slices"
+ n_views = views_ind.shape[0]-1
+ else:
+ if views_ind.ndim == 1:
+ if not views_ind.dtype == np.object:
+ raise ValueError("The format of views_ind is not "
+ + "supported.")
+ for ind, val in enumerate(views_ind):
+ views_ind[ind] = np.array(val)
+ if not np.issubdtype(views_ind[ind].dtype, np.integer):
+ raise ValueError("Values in views_ind must be "
+ + "integers.")
+ if views_ind[ind].min() < 0 \
+ or views_ind[ind].max() >= n_features:
+ raise ValueError("Values in views_ind are not in a "
+ + "correct range for the provided "
+ + "data.")
+ elif views_ind.ndim == 2:
+ if not np.issubdtype(views_ind.dtype, np.integer):
+ raise ValueError("Values in views_ind must be integers.")
+ if views_ind.min() < 0 or views_ind.max() >= n_features:
+ raise ValueError("Values in views_ind are not in a "
+ + "correct range for the provided data.")
+ else:
+ raise ValueError("The format of views_ind is not supported.")
+ self.view_mode_ = "indices"
+ n_views = views_ind.shape[0]
+ self.views_ind = views_ind
+ self.n_views = n_views
+ return (views_ind, n_views)
+
+class MultiModalSparseInfo():
+
+ def __init__(self, data, view_ind=None):
+ """Constructor of Metriclearn_array"""
+ shapes_int = []
+ index = 0
+ new_data = np.ndarray([])
+ n_views = data.size
+ thekeys = None
+ # view_ind_self = None
+ view_mode = 'slices'
+
+ if (sp.issparse(data)) and data.ndim > 1:
+ if view_ind is not None:
+ try:
+ view_ind = np.asarray(view_ind)
+ except :
+ raise TypeError("n_views should be list or nparray")
+ elif view_ind is None:
+ if data.shape[1] > 1:
+ view_ind = np.array([0, data.shape[1]//2, data.shape[1]])
+ else:
+ view_ind = np.array([0, data.shape[1]])
+
+ new_data = data
+ # view_ind_self = view_ind
+ view_ind, n_views, view_mode = self._first_validate_views_ind(view_ind,
+ data.shape[1])
+ if view_ind.ndim == 1 and view_mode.startswith("slicing"):
+ shapes_int = [in2 - in1 for in1, in2 in zip(view_ind, view_ind[1:])]
+
+ if data.shape[0] < 1 or data.shape[1] < 1:
+ raise ValueError("input data shouldbe not empty")
+ self.view_mode_ = view_mode
+ self.views_ind = view_ind
+ self.shapes_int = shapes_int
+ self.n_views = n_views
+
+
+class MultiModalSparseArray(sp.csr_matrix, sp.csc_matrix, MultiModalSparseInfo, MultiModalData):
+ """
+ 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
+ ----------
+
+ view_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))
+
+ """
+
+ def __init__(self, *arg, **kwargs ):
+ """Constructor of Metriclearn_array"""
+ if sp.issparse(arg[0]):
+ MultiModalSparseInfo.__init__(self, *arg)
+ if isinstance(arg[0], sp.csr_matrix) :
+ sp.csr_matrix.__init__(self, arg[0])
+ elif isinstance(arg[0], sp.csc_matrix):
+ sp.csc_matrix.__init__(self, arg[0])
+ else:
+ raise TypeError("This sparse format is not supported")
+ else:
+ if isinstance(self,sp.csr_matrix):
+ sp.csr_matrix.__init__(self, *arg, **kwargs)
+ elif isinstance(self, sp.csc_matrix):
+ sp.csc_matrix.__init__(self, *arg, **kwargs)
+
+
+
+
+if __name__ == '__main__':
+ rng = np.random.RandomState(0)
+ X = rng.rand(40, 10)
+ X[X < .8] = 0
+ X_csr = sp.csr_matrix(X)
+ y = (4 * rng.rand(40)).astype(np.int)
+ X_ = MultiModalSparseArray(X_csr)
+ print(X_.shape)
+ print(X_[:,0:1])
+
+ X = np.array([[3, 0], [0, 1], [0, 2], [1, 1], [1, 2], [2, 1]])
+ y = [1, 1, 1, 2, 2, 2]
+ clf = MumboClassifier()
+ clf.fit(X, y)
\ No newline at end of file
diff --git a/multimodal/tests/test_cumbo.py b/multimodal/tests/test_cumbo.py
index d8a9e00..153bc94 100644
--- a/multimodal/tests/test_cumbo.py
+++ b/multimodal/tests/test_cumbo.py
@@ -17,7 +17,7 @@ from sklearn import datasets
from multimodal.boosting.cumbo import MuCumboClassifier
from multimodal.tests.data.get_dataset_path import get_dataset_path
-
+from multimodal.datasets.data_sample import MultiModalArray
class TestMuCumboClassifier(unittest.TestCase):
@@ -909,11 +909,11 @@ class TestMuCumboClassifier(unittest.TestCase):
(self.iris.data, target_two_classes, self.iris.views_ind),
(self.iris.data, target_two_classes, np.array([[0, 2], [1, 3]])),
)
+
#
for X, y, views_ind in data:
clf = MuCumboClassifier(n_estimators=n_estimators, random_state=seed)
clf.fit(X, y, views_ind)
-
staged_dec_func = [dec_f for dec_f in clf.staged_decision_function(X)]
staged_predict = [predict for predict in clf.staged_predict(X)]
staged_score = [score for score in clf.staged_score(X, y)]
diff --git a/multimodal/tests/test_data_sample.py b/multimodal/tests/test_data_sample.py
index 43e3540..04c6b49 100644
--- a/multimodal/tests/test_data_sample.py
+++ b/multimodal/tests/test_data_sample.py
@@ -1,9 +1,9 @@
import unittest
import numpy as np
-from metriclearning.datasets.base import load_dict
-from metriclearning.tests.datasets.get_dataset_path import get_dataset_path
-from metriclearning.datasets.data_sample import Metriclearn_array
+from multimodal.datasets.base import load_dict
+from multimodal.tests.datasets.get_dataset_path import get_dataset_path
+from multimodal.datasets.data_sample import MultiModalArray
import pickle
class UnitaryTest(unittest.TestCase):
@@ -29,12 +29,12 @@ class UnitaryTest(unittest.TestCase):
def testGet_view(self):
- a = Metriclearn_array(self.kernel_dict)
+ a = MultiModalArray(self.kernel_dict)
np.testing.assert_almost_equal(a.get_view(0), self.kernel_dict[0], 8)
np.testing.assert_almost_equal(a.get_view(1), self.kernel_dict[1], 8)
def test_init_Metriclearn_array(self):
- a = Metriclearn_array(self.kernel_dict)
+ a = MultiModalArray(self.kernel_dict)
self.assertEqual(a.shape, (120, 240))
self.assertEqual(a.shapes_int, [120, 120])
self.assertEqual(a.n_views, 2)
@@ -42,9 +42,9 @@ class UnitaryTest(unittest.TestCase):
self.assertEqual(a.keys, dict_key.keys())
def test_init_Array(self):
- a = Metriclearn_array(self.kernel_dict)
+ a = MultiModalArray(self.kernel_dict)
array_x = a.data
- b = Metriclearn_array(a)
+ b = MultiModalArray(a)
np.testing.assert_equal(b.views_ind, np.array([0, 120, 240]))
diff --git a/multimodal/tests/test_mkl.py b/multimodal/tests/test_mkl.py
index fffbac5..16008f9 100644
--- a/multimodal/tests/test_mkl.py
+++ b/multimodal/tests/test_mkl.py
@@ -3,9 +3,9 @@
import unittest
import numpy as np
from sklearn.metrics.pairwise import rbf_kernel
-from metriclearning.tests.datasets.get_dataset_path import get_dataset_path
-from metriclearning.lpMKL import MKL
-from metriclearning.datasets.data_sample import Metriclearn_array
+from multimodal.tests.datasets.get_dataset_path import get_dataset_path
+from multimodal.kernels.lpMKL import MKL
+from multimodal.datasets.data_sample import MultiModalArray
import pickle
from sklearn.exceptions import NotFittedError
@@ -70,7 +70,7 @@ class MKLTest(unittest.TestCase):
#######################################################
# mvml = MVML.fit(self.kernel_dict, self.y)
w_expected = np.array([[0.5], [0.5]])
- x_metricl = Metriclearn_array(self.kernel_dict)
+ x_metricl = MultiModalArray(self.kernel_dict)
mkl2 = MKL(lmbda=3, m_param = 0.3, kernel=['rbf'], kernel_params=[{'gamma':50}],
use_approx = True,
precision = 1E0, n_loops = 50)
@@ -83,7 +83,7 @@ class MKLTest(unittest.TestCase):
#######################################################
# mvml = MVML.fit(self.kernel_dict, self.y)
w_expected = np.array([[0.5], [0.5]])
- x_metricl = Metriclearn_array(self.kernel_dict)
+ x_metricl = MultiModalArray(self.kernel_dict)
mkl2 = MKL(lmbda=3, m_param = 0.3, kernel="precomputed",
use_approx = True,
precision = 1E-9, n_loops = 600)
@@ -97,7 +97,7 @@ class MKLTest(unittest.TestCase):
mkl.predict(self.test_kernel_dict)
def testPredictMVML_witoutFit(self):
- x_metric = Metriclearn_array(self.kernel_dict)
+ x_metric = MultiModalArray(self.kernel_dict)
mkl = MKL(lmbda=3, m_param = 0.3, kernel=['rbf'], kernel_params=[{'gamma':50}],
use_approx = True,
precision = 1E-9, n_loops = 50)
diff --git a/multimodal/tests/test_mumbo.py b/multimodal/tests/test_mumbo.py
index f5b3924..978244f 100644
--- a/multimodal/tests/test_mumbo.py
+++ b/multimodal/tests/test_mumbo.py
@@ -35,7 +35,7 @@ from sklearn.ensemble import RandomForestClassifier
from sklearn.cluster import KMeans
from sklearn.tree import DecisionTreeClassifier
from sklearn import datasets
-from multimodalboost.mumbo import MumboClassifier
+from multimodal.boosting.mumbo import MumboClassifier
class TestMuCumboClassifier(unittest.TestCase):
@@ -47,6 +47,15 @@ class TestMuCumboClassifier(unittest.TestCase):
iris.views_ind = np.array([0, 2, 4])
clf.iris = iris
+ def test_sparse(self):
+ rng = np.random.RandomState(0)
+ X = rng.rand(40, 10)
+ X[X < .8] = 0
+ X_csr = csr_matrix(X)
+ clf = MumboClassifier()
+ y = (4 * rng.rand(40)).astype(np.int)
+ clf.fit(X_csr, y)
+
def test_init_var(self):
n_classes = 3
@@ -318,7 +327,7 @@ class TestMuCumboClassifier(unittest.TestCase):
np.random.seed(seed)
n_estimators = 10
-
+ #print("iris views ind", self.iris.views_ind)
clf = MumboClassifier(n_estimators=n_estimators, best_view_mode='edge')
clf.fit(self.iris.data, self.iris.target, self.iris.views_ind)
score = clf.score(self.iris.data, self.iris.target)
@@ -347,7 +356,7 @@ class TestMuCumboClassifier(unittest.TestCase):
expected_views_ind = np.array([0, 1, 3])
clf = MumboClassifier()
clf.fit(X, y)
- np.testing.assert_equal(clf.views_ind_, expected_views_ind)
+ np.testing.assert_equal(clf.X_.views_ind, expected_views_ind)
# Check that classes labels can be integers or strings and can be stored
# into any kind of sequence
@@ -515,6 +524,7 @@ class TestMuCumboClassifier(unittest.TestCase):
np.testing.assert_equal(clf.predict(X), y)
np.testing.assert_equal(clf.predict(np.array([[1., 1.], [-1., -1.]])),
np.array([0, 1]))
+ X = clf._global_X_transform(X, clf.X_.views_ind)
self.assertEqual(clf.decision_function(X).shape, y.shape)
views_ind = np.array([[1, 0]])
@@ -695,6 +705,11 @@ class TestMuCumboClassifier(unittest.TestCase):
def test_classifier(self):
+ X_zero_features = np.empty(0).reshape(3, 0)
+ y = np.array([1, 0, 1])
+ # e = MumboClassifier()
+ # e.fit(X_zero_features, y)
+ # print(e.predict(X_zero_features))
return check_estimator(MumboClassifier)
def test_iris(self):
@@ -742,7 +757,6 @@ class TestMuCumboClassifier(unittest.TestCase):
for X, y, views_ind in data:
clf = MumboClassifier(n_estimators=n_estimators, random_state=seed)
clf.fit(X, y, views_ind)
-
staged_dec_func = [dec_f for dec_f in clf.staged_decision_function(X)]
staged_predict = [predict for predict in clf.staged_predict(X)]
staged_score = [score for score in clf.staged_score(X, y)]
@@ -782,7 +796,6 @@ class TestMuCumboClassifier(unittest.TestCase):
clf.fit(self.iris.data, self.iris.target, self.iris.views_ind)
score = clf.score(self.iris.data, self.iris.target)
dump = pickle.dumps(clf)
-
clf_loaded = pickle.loads(dump)
self.assertEqual(type(clf_loaded), clf.__class__)
score_loaded = clf_loaded.score(self.iris.data, self.iris.target)
@@ -828,11 +841,9 @@ class TestMuCumboClassifier(unittest.TestCase):
X_dense = self.iris.data
y = self.iris.target
- for sparse_format in [csc_matrix, csr_matrix, lil_matrix, coo_matrix,
- dok_matrix]:
+ for sparse_format in [csc_matrix, csr_matrix]: #, lil_matrix, coo_matrix,dok_matrix]:
for views_ind in (self.iris.views_ind, np.array([[0, 2], [1, 3]])):
X_sparse = sparse_format(X_dense)
-
clf_sparse = MumboClassifier(
base_estimator=CustomSVC(),
random_state=seed,
@@ -872,9 +883,9 @@ class TestMuCumboClassifier(unittest.TestCase):
# Check that sparsity of data is maintained during training
types = [clf.data_type_ for clf in clf_sparse.estimators_]
if sparse_format == csc_matrix:
- self.assertTrue(all([type_ == csc_matrix for type_ in types]))
+ self.assertTrue(all([issubclass(type_, csc_matrix) for type_ in types]))
else:
- self.assertTrue(all([type_ == csr_matrix for type_ in types]))
+ self.assertTrue(all([issubclass(type_, csr_matrix) for type_ in types]))
if __name__ == '__main__':
diff --git a/multimodal/tests/test_mvml.py b/multimodal/tests/test_mvml.py
index 3de8a35..4627c5d 100644
--- a/multimodal/tests/test_mvml.py
+++ b/multimodal/tests/test_mvml.py
@@ -6,9 +6,9 @@ import unittest
import numpy as np
from sklearn.exceptions import NotFittedError
-from metriclearning.datasets.data_sample import Metriclearn_array
-from metriclearning.mvml import MVML
-from metriclearning.tests.datasets.get_dataset_path import get_dataset_path
+from multimodal.datasets.data_sample import MultiModalArray
+from multimodal.kernels.mvml import MVML
+from multimodal.tests.datasets.get_dataset_path import get_dataset_path
class MVMLTest(unittest.TestCase):
@@ -90,7 +90,7 @@ class MVMLTest(unittest.TestCase):
#######################################################
# mvml = MVML.fit(self.kernel_dict, self.y)
w_expected = np.array([[0.5], [0.5]])
- x_metricl = Metriclearn_array(self.kernel_dict)
+ x_metricl = MultiModalArray(self.kernel_dict)
mvml2 = MVML(lmbda=0.1, eta=1, nystrom_param=1.0)
mvml2.fit(x_metricl, y=self.y, views_ind=None)
self.assertEqual(mvml2.A.shape, (240, 240))
@@ -105,7 +105,7 @@ class MVMLTest(unittest.TestCase):
#######################################################
# mvml = MVML.fit(self.kernel_dict, self.y)
w_expected = np.array([[0.5], [0.5]])
- x_metricl = Metriclearn_array(self.kernel_dict)
+ x_metricl = MultiModalArray(self.kernel_dict)
mvml2 = MVML(lmbda=0.1, eta=1, nystrom_param=1.0, learn_A=4)
mvml2.fit(x_metricl, y=self.y, views_ind=None)
self.assertEqual(mvml2.A.shape, (240, 240))
@@ -120,7 +120,7 @@ class MVMLTest(unittest.TestCase):
#######################################################
# mvml = MVML.fit(self.kernel_dict, self.y)
w_expected = np.array([[0.5], [0.5]])
- x_metricl = Metriclearn_array(self.kernel_dict)
+ x_metricl = MultiModalArray(self.kernel_dict)
mvml2 = MVML(lmbda=0.1, eta=1, nystrom_param=1.0, learn_A=3)
mvml2.fit(x_metricl, y=self.y, views_ind=None)
self.assertEqual(mvml2.A.shape, (240, 240))
@@ -134,7 +134,7 @@ class MVMLTest(unittest.TestCase):
# task with Metric array
#######################################################
w_expected = np.array([0.2, 0.1]) # [0.94836083 , 0.94175933] [ 0.7182, 0.7388]
- x_metricl = Metriclearn_array(self.kernel_dict)
+ x_metricl = MultiModalArray(self.kernel_dict)
mvml2 = MVML(lmbda=0.1, eta=1, nystrom_param=0.6,
learn_A=2, learn_w=1)
mvml2.fit(x_metricl, y=self.y, views_ind=None)
@@ -149,7 +149,7 @@ class MVMLTest(unittest.TestCase):
# task with Metric array
#######################################################
w_expected = np.array([1.3, 1.4]) # [0.94836083 , 0.94175933] [ 0.7182, 0.7388]
- x_metricl = Metriclearn_array(self.kernel_dict)
+ x_metricl = MultiModalArray(self.kernel_dict)
mvml2 = MVML(lmbda=0.1, eta=1, nystrom_param=0.6,
learn_A=1, learn_w=1)
mvml2.fit(x_metricl, y=self.y, views_ind=None)
@@ -164,7 +164,7 @@ class MVMLTest(unittest.TestCase):
# task with nparray 2d
#######################################################
w_expected = np.array([[0.5], [0.5]])
- x_metricl = Metriclearn_array(self.kernel_dict)
+ x_metricl = MultiModalArray(self.kernel_dict)
x_array = np.asarray(x_metricl)
mvml3 = MVML(lmbda=0.1, eta=1, nystrom_param=1.0)
mvml3.fit(x_array, y=self.y, views_ind=[0, 120, 240])
--
GitLab