def test_sentiment_analysis(no_samples, sess):
data_X, data_Y, cv_X, cv_Y, test_X, test_Y = get_dataset(all_globals.dataset_name, all_globals.is_binarized, all_globals.is_resized, all_globals.is_grayscale)
no_features = data_X.shape[1]
print(data_X.shape, data_Y.shape)
data_X, data_Y = data_X[:no_samples[0]], data_Y[:no_samples[0]]
##print(data_X.shape, data_Y.shape)
print(np.unique(data_Y))
interpret_train = []
interpret_test = []
interpret_cv = []
for itera in range(3):
model_B = ModelB(sess, all_globals.model_B_epochs, all_globals.batch_size, all_globals.dataset_name, all_globals.learning_rate, all_globals.model_name_B, all_globals.is_binarized, all_globals.is_resized, all_globals.is_grayscale, no_features)
model_A = ModelA(sess, all_globals.model_A_epochs, all_globals.batch_size, all_globals.dataset_name, all_globals.learning_rate, all_globals.model_name_A, all_globals.is_binarized, all_globals.is_resized, all_globals.is_grayscale, no_features)
model_B.init_model()
#X_train, X_cross_validation, Y_train, Y_cross_validation = train_test_split(data_X, data_Y, test_size = 0.2, random_state = 42, stratify = data_Y)
print("Iteration " + str(itera))
model_B.set_dataset(data_X, data_Y, test_X, test_Y, cv_X, cv_Y)
model_B.train_model()
##print("Model B trained")
model_A.set_dataset(data_X, data_Y, test_X, test_Y, cv_X, cv_Y)
outp_train, outp_cv, outp_test = [], [], []
outp_train, outp_cv, outp_test = perform_interpretation(all_globals.mc_dropout, outp_train, outp_cv, outp_test, model_A, model_B)
print("Train", find_average(outp_train))
print("CV", find_average(outp_cv))
print("Test", find_average(outp_test))
interpret_train.append(find_average(outp_train))
interpret_test.append(find_average(outp_test))
interpret_cv.append(find_average(outp_cv))
#tf.compat.v1.reset_default_graph() <-------- THIS DOESN'T WORK, gives error: AssertionError: Do not use tf.reset_default_graph() to clear nested graphs. If you need a cleared graph, exit the nesting and create a new graph.
print("Final Interpretability on CV", find_average(interpret_cv))
print("Final interpretability on Train", find_average(interpret_train))
print("Final interpretability on Test", find_average(interpret_test))
def test_kfold_cross_validation_stanford40():
data_X, data_Y, test_X, test_Y, data_X_A, test_X_A = get_dataset(
'stanford40', all_globals.is_binarized, all_globals.is_resized,
all_globals.is_grayscale)
interpret_train = []
interpret_test = []
interpret_cv = []
for itera in range(3):
kf = KFold(n_splits=all_globals.no_of_folds,
shuffle=True,
random_state=None)
outp_train, outp_cv, outp_test = [], [], []
indices1 = list(kf.split(data_X))
for i in range(len(indices1)):
model_B = ModelB(all_globals.model_B_epochs,
all_globals.batch_size, all_globals.dataset_name,
all_globals.learning_rate,
all_globals.model_name_B,
all_globals.is_binarized, all_globals.is_resized,
all_globals.is_grayscale)
model_A = ModelA(all_globals.model_A_epochs,
all_globals.batch_size, all_globals.dataset_name,
all_globals.learning_rate,
all_globals.model_name_A,
all_globals.is_binarized, all_globals.is_resized,
all_globals.is_grayscale)
model_B.init_model()
print("Cross Validation fold " + str(i + 1))
#print("TRAIN INDEXES:", train_index, "CROSS_VALIDATION INDEXES:", cross_validation_index)
train_index, cross_validation_index = indices1[i]
X_train, X_cross_validation = data_X[train_index], data_X[
cross_validation_index]
Y_train, Y_cross_validation = data_Y[train_index], data_Y[
cross_validation_index]
model_B.set_dataset(X_train, Y_train, test_X, test_Y,
X_cross_validation, Y_cross_validation)
model_B.train_model()
##print("Model B trained")
train_index, cross_validation_index = indices1[i]
X_train_A, X_cross_validation_A = data_X_A[train_index], data_X_A[
cross_validation_index]
Y_train_A, Y_cross_validation_A = data_Y[train_index], data_Y[
cross_validation_index]
model_A.set_dataset(X_train_A, Y_train_A, test_X_A, test_Y,
X_cross_validation_A, Y_cross_validation_A)
outp_train, outp_cv, outp_test = perform_interpretation(
all_globals.mc_dropout, outp_train, outp_cv, outp_test,
model_A, model_B)
print("Train", find_average(outp_train))
print("CV", find_average(outp_cv))
print("Test", find_average(outp_test))
interpret_train.append(find_average(outp_train))
interpret_test.append(find_average(outp_test))
interpret_cv.append(find_average(outp_cv))
print("Final Interpretability on CV", find_average(interpret_cv))
print("Final interpretability on Train", find_average(interpret_train))
print("Final interpretability on Test", find_average(interpret_test))
def test_kfold_cross_validation(no_samples, sess):
data_X, data_Y, test_X, test_Y = get_dataset(all_globals.dataset_name, all_globals.is_binarized, all_globals.is_resized, all_globals.is_grayscale)
print(data_X.shape, data_Y.shape)
data_X, data_Y = data_X[:no_samples[0]], data_Y[:no_samples[0]]
##print(data_X.shape, data_Y.shape)
print(np.unique(data_Y))
final_train, final_cv, final_test = [], [], []
interpret_train = []
interpret_test = []
interpret_cv = []
for itera in range(2):
print("Iteration" + str(itera))
kf = KFold(n_splits=all_globals.no_of_folds, shuffle = True, random_state = None)
i = 1
outp_train, outp_cv, outp_test = [], [], []
for train_index, cross_validation_index in kf.split(data_X):
print("Cross Validation fold " + str(i))
#print("TRAIN INDEXES:", train_index, "CROSS_VALIDATION INDEXES:", cross_validation_index)
model_B = ModelB(sess, all_globals.model_B_epochs, all_globals.batch_size, all_globals.dataset_name, all_globals.learning_rate, all_globals.model_name_B, all_globals.is_binarized, all_globals.is_resized, all_globals.is_grayscale)
model_A = ModelA(sess, all_globals.model_A_epochs, all_globals.batch_size, all_globals.dataset_name, all_globals.learning_rate, all_globals.model_name_A, all_globals.is_binarized, all_globals.is_resized, all_globals.is_grayscale)
model_B.init_model()
X_train, X_cross_validation = data_X[train_index], data_X[cross_validation_index]
Y_train, Y_cross_validation = data_Y[train_index], data_Y[cross_validation_index]
model_B.set_dataset(X_train, Y_train, test_X, test_Y, X_cross_validation, Y_cross_validation)
model_B.train_model()
##print("Model B trained")
model_A.set_dataset(X_train, Y_train, test_X, test_Y, X_cross_validation, Y_cross_validation)
outp_train, outp_cv, outp_test = perform_interpretation(all_globals.mc_dropout, outp_train, outp_cv, outp_test, model_A, model_B)
i = i + 1
print("Train ", find_average(outp_train))
print("CV ", find_average(outp_cv))
print("Test ", find_average(outp_test))
final_train.extend(outp_train)
final_cv.extend(outp_cv)
final_test.extend(outp_test)
final_train.append(find_average(outp_train))
final_cv.append(find_average(outp_cv))
final_test.append(find_average(outp_test))
#print_to_file(find_average(outp_train), find_average(outp_cv), find_average(outp_test))
interpret_train.append(find_average(outp_train))
interpret_test.append(find_average(outp_test))
interpret_cv.append(find_average(outp_cv))
print("Final Interpretability on CV", find_average(interpret_cv))
print("Final interpretability on Train", find_average(interpret_train))
print("Final interpretability on Test", find_average(interpret_test))
final_train.append(find_average(interpret_train))
final_cv.append(find_average(interpret_cv))
final_test.append(find_average(interpret_test))
print(final_train)
print(final_cv)
print(final_test)
def test_cross_validation(no_samples):
data_X, data_Y, test_X, test_Y = get_dataset(all_globals.dataset_name,
all_globals.is_binarized,
all_globals.is_resized,
all_globals.is_grayscale)
data_X, data_Y = data_X[:no_samples[0]], data_Y[:no_samples[0]]
interpret_train = []
interpret_test = []
interpret_cv = []
for itera in range(3):
model_B = ModelB(all_globals.model_B_epochs, all_globals.batch_size,
all_globals.dataset_name, all_globals.learning_rate,
all_globals.model_name_B, all_globals.is_binarized,
all_globals.is_resized, all_globals.is_grayscale)
model_A = ModelA(all_globals.model_A_epochs, all_globals.batch_size,
all_globals.dataset_name, all_globals.learning_rate,
all_globals.model_name_A, all_globals.is_binarized,
all_globals.is_resized, all_globals.is_grayscale)
model_B.init_model()
X_train, X_cross_validation, Y_train, Y_cross_validation = train_test_split(
data_X, data_Y, test_size=0.2, random_state=42, stratify=data_Y)
print("Iteration " + str(itera))
model_B.set_dataset(X_train, Y_train, test_X, test_Y,
X_cross_validation, Y_cross_validation)
model_B.train_model()
##print("Model B trained")
model_A.set_dataset(X_train, Y_train, test_X, test_Y,
X_cross_validation, Y_cross_validation)
outp_train, outp_cv, outp_test = [], [], []
outp_train, outp_cv, outp_test = perform_interpretation(
all_globals.mc_dropout, outp_train, outp_cv, outp_test, model_A,
model_B)
print("Train", find_average(outp_train))
print("CV", find_average(outp_cv))
print("Test", find_average(outp_test))
interpret_train.append(find_average(outp_train))
interpret_test.append(find_average(outp_test))
interpret_cv.append(find_average(outp_cv))
print("Final Interpretability on CV", find_average(interpret_cv))
print("Final interpretability on Train", find_average(interpret_train))
print("Final interpretability on Test", find_average(interpret_test))
def interpretation_diff_frequencies(no_samples, sess):
frequencies_list_width = [10]
data_X, data_Y, test_X, test_Y = get_dataset(all_globals.dataset_name, all_globals.is_binarized, all_globals.is_resized, all_globals.is_grayscale)
data_X, data_Y = data_X[:no_samples[0]], data_Y[:no_samples[0]]
for width in frequencies_list_width:
print("Width: " + str(width) + "\n")
high_interpret_train, high_interpret_test, high_interpret_cv = [], [], []
low_interpret_train, low_interpret_test, low_interpret_cv = [], [], []
for itera in range(3):
kf = KFold(n_splits=all_globals.no_of_folds, shuffle = True)
i = 1
high_outp_train, high_outp_cv, high_outp_test = [], [], []
low_outp_train, low_outp_cv, low_outp_test = [], [], []
for train_index, cross_validation_index in kf.split(data_X):
model_B = ModelB(sess, all_globals.model_B_epochs, all_globals.batch_size, all_globals.dataset_name, all_globals.learning_rate, all_globals.model_name_B, all_globals.is_binarized, all_globals.is_resized, all_globals.is_grayscale)
model_A = ModelA(sess, all_globals.model_A_epochs, all_globals.batch_size, all_globals.dataset_name, all_globals.learning_rate, all_globals.model_name_A, all_globals.is_binarized, all_globals.is_resized, all_globals.is_grayscale)
model_B.init_model()
print("Cross Validation fold " + str(i))
X_train, X_cross_validation = data_X[train_index], data_X[cross_validation_index]
Y_train, Y_cross_validation = data_Y[train_index], data_Y[cross_validation_index]
high_freq_train, low_freq_train = get_frequency_components_dataset(X_train, width)
high_freq_test, low_freq_test = get_frequency_components_dataset(test_X, width)
high_freq_cv, low_freq_cv = get_frequency_components_dataset(X_cross_validation, width)
model_B.set_dataset(X_train, Y_train, test_X, test_Y, X_cross_validation, Y_cross_validation)
model_B.train_model()
##print("Model B trained")
model_A.set_dataset(high_freq_train, Y_train, high_freq_test, test_Y, high_freq_cv, Y_cross_validation)
high_outp_train, high_outp_cv, high_outp_test = perform_interpretation(all_globals.mc_dropout, high_outp_train, high_outp_cv, high_outp_test)
model_A.set_dataset(low_freq_train, Y_train, low_freq_test, test_Y, low_freq_cv, Y_cross_validation)
low_outp_train, low_outp_cv, low_outp_test = perform_interpretation(all_globals.mc_dropout, low_outp_train, low_outp_cv, low_outp_test, model_A, model_B)
i = i + 1
print("High_Freq:\nTrain: " + str(find_average(high_outp_train)) + "\nCV: " + str(find_average(high_outp_cv)) + "\nTest: " + str(find_average(high_outp_test)))
print("Low_Freq:\nTrain: " + str(find_average(low_outp_train)) + "\nCV: " + str(find_average(low_outp_cv)) + "\nTest: " + str(find_average(low_outp_test)))
high_interpret_train.append(find_average(high_outp_train))
high_interpret_test.append(find_average(high_outp_test))
high_interpret_cv.append(find_average(high_outp_cv))
low_interpret_train.append(find_average(low_outp_train))
low_interpret_test.append(find_average(low_outp_test))
low_interpret_cv.append(find_average(low_outp_cv))
print("Final Values\nTrain: " + str(find_average(high_interpret_train)) + "\nCV: " + str(find_average(high_interpret_cv)) + "\nTest: " + str(find_average(high_interpret_test)))
print("Final Values\nTrain: " + str(find_average(low_interpret_train)) + "\nCV: " + str(find_average(low_interpret_cv)) + "\nTest: " + str(find_average(low_interpret_test)))