def extraction(self, data):
# extracting 40 features using sparse filtering
sp = SparseFiltering(self.num_feature_extract)
feature_set = sp.fit_transform(data)
# fig = plt.figure()
# ax = fig.add_subplot(111, projection='3d')
# ax.scatter(feature_set[:, 0], feature_set[:, 1], feature_set[:, 2], c=label[:, 0])
# ax.set_xlabel('feature 1')
# ax.set_ylabel('feature 2')
# ax.set_zlabel('feature 3')
# plt.title('SP visulization')
# plt.show()
return feature_set
pred = clf.predict(XXtrainIn)
#save the classifier
_ = joblib.dump(clf, 'random_forest_trained.pkl', compress =9)
labels = ytrain.squeeze()
report_accuracy(pred,labels, ' RF Raw Train')
pred = clf.predict(XXtest)
labels = ytest.squeeze()
report_accuracy(pred,labels, ' RF Raw Test')
#SparseFiltering
n_features = 64 # How many features are learned
estimator = SparseFiltering(n_features=n_features,
maxfun=200, # The maximal number of evaluations of the objective function
iprint=-1) # after how many function evaluations is information printed
# by L-BFGS. -1 for no information
print("sparse yapıldı")
#model2 = RandomForestClassifier(n_estimators=20, max_depth=100, criterion='entropy', max_features='sqrt',max_leaf_nodes=30,oob_score=False)
train_features = estimator.fit_transform(XXtrainIn)
rfr = RandomForestClassifier(n_estimators=51, class_weight=wtrain, max_depth=15)
rfr.fit(train_features,ytrain.ravel())
pred = rfr.predict(train_features)
report_accuracy(pred,ytrain.ravel(), 'Sparse RF Train')
test_features = estimator.transform(XXtest)
pred = rfr.predict(test_features)
report_accuracy(pred,ytest.ravel(), 'Sparse RF Test')
faces_centered = \
faces_centered.reshape(n_samples, 64, 64) # Reshaping to 64*64 pixel images
print("Dataset consists of %d faces" % n_samples)
###############################################################################
# Extract n_patches patches randomly from each image
patches = [extract_patches_2d(faces_centered[i], (patch_width, patch_width),
max_patches=n_patches, random_state=i)
for i in range(n_samples)]
patches = np.array(patches).reshape(-1, patch_width * patch_width)
###############################################################################
#
estimator = SparseFiltering(n_features=n_features, maxfun=maxfun, iprint=iprint)
features = estimator.fit_transform(patches)
# #############################################################################
# Plot weights of features
pl.figure(0, figsize=(12, 10))
pl.subplots_adjust(left=0.01, bottom=0.01, right=0.99, top=0.95,
wspace=0.1, hspace=0.4)
for i in range(estimator.w_.shape[0]):
pl.subplot(int(np.sqrt(n_features)), int(np.sqrt(n_features)), i + 1)
pl.pcolor(estimator.w_[i].reshape(patch_width, patch_width),
cmap=pl.cm.gray)
pl.xticks(())
pl.yticks(())
pl.title("Feature %4d" % i)
#
# row_sums = test_X.sum(axis=1).astype(float)
# test_X = np.true_divide(test_X, row_sums[:, np.newaxis])
###############################################################################
# Dictionary Learning
n_components = 700
n_samples_training = 20000
print("\nSparse Coding Dictionary Learning")
# pca = RandomizedPCA(n_components=n_dcomponents).fit(train_X)
# dl = KSVDSparseCoding(n_components, n_nonzero_coefs=70, preserve_dc=False, approx=False, max_iter=5, verbose=1)
# dl.fit(train_X[0:n_samples_training])
estimator = SparseFiltering(n_features=n_components,
maxfun=500, # The maximal number of evaluations of the objective function
iprint=10) # after how many function evaluations is information printed
# by L-BFGS. -1 for no information
features = estimator.fit_transform(train_X[0:n_samples_training])
print "features.shape", features.shape
print "estimator.w_.shape", estimator.w_.shape
plt.figure(figsize=(12, 10))
for i in range(estimator.w_.shape[0]):
plt.subplot(int(np.sqrt(n_features)), int(np.sqrt(n_features)), i + 1)
plt.pcolor(estimator.w_[i].reshape(w, h),
cmap=plt.cm.gray, vmin=estimator.w_.min(),
vmax=estimator.w_.max())
plt.xticks(())
plt.yticks(())
plt.title("")
faces_centered = \
faces_centered.reshape(n_samples, 64, 64) # Reshaping to 64*64 pixel
print("Dataset consists of %d faces" % n_samples)
###############################################################################
# Extract n_patches patches randomly from each image
patches = [
extract_patches_2d(faces_centered[i], (patch_width, patch_width),
max_patches=n_patches,
random_state=i) for i in range(n_samples)
]
patches = np.array(patches).reshape(-1, patch_width * patch_width)
###############################################################################
estimator = \
SparseFiltering(n_features=n_features, maxfun=maxfun, iprint=iprint)
features = estimator.fit_transform(patches)
# #############################################################################
# Plot weights of features
pl.figure(0, figsize=(12, 10))
pl.subplots_adjust(left=0.01,
bottom=0.01,
right=0.99,
top=0.95,
wspace=0.1,
hspace=0.4)
for i in range(estimator.w_.shape[0]):
pl.subplot(int(np.sqrt(n_features)), int(np.sqrt(n_features)), i + 1)
pl.pcolor(estimator.w_[i].reshape(patch_width, patch_width),
cmap=pl.cm.gray)
