def pointwise_mutual_information_score(X, y):
#TODO Should binarize things
#Filter negative mutual information
X = check_array(X, accept_sparse='csr')
if np.any((X.data if issparse(X) else X) < 0):
raise ValueError("Input X must be non-negative.")
if issparse(X):
X_bin = sp.csr_matrix(X)
X_bin.data = np.ones(X_bin.data.shape)
else:
raise ValueError('Matrix should be sparse')
Y = LabelBinarizer().fit_transform(y)
if Y.shape[1] == 1:
Y = np.append(1 - Y, Y, axis=1)
observed = safe_sparse_dot(Y.T, X_bin,
dense_output=True) # n_classes * n_features
feature_count = X_bin.sum(axis=0).reshape(1, -1)
n_feat = X_bin.shape[1]
n_classes = Y.shape[1] #Should be 2 for binary classes
class_count = Y.sum(axis=0).reshape(1, -1)
F = np.tile(feature_count, (n_classes, 1))
C = np.tile(class_count, (n_feat, 1))
#embed()
print(feature_count.shape)
print(class_count.shape)
eps = 1e-8
#Do real Mutual Information
PMI = np.log(eps + observed) - np.log(C.T) - np.log(F)
PMI = np.maximum(PMI, 0)
return np.asarray(PMI.sum(axis=0)).squeeze()
# initialize the image preprocessors
aap = AspectAwarePreprocessor(224, 224, gray=True)
iap = ImageToArrayPreprocessor()
# load the dataset from disk then scale the raw pixel intensities
# to the range [0, 1]
sdl = SimpleDatasetLoader(preprocessor=[iap], gray=True)
data, labels = sdl.load(imagePaths, verbose=500)
data = data / 255
# # convert the labels from integers to vectors
labels = LabelBinarizer().fit_transform(labels)
# account for skew in the labeled data
classTotals = labels.sum(axis=0)
classWeight = classTotals.max() / classTotals
# partition the data into training and testing splits using 75% of
# the data for training and the remaining 25% for testing
trainX, testX, trainY, testY = train_test_split(data,
labels,
test_size=0.25,
stratify=labels,
random_state=42)
# construct the image generator for data augmentation
aug = ImageDataGenerator(rotation_range=30,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=0.2,