def experiment(sparseness, index, percentage, extend_near_num, matrix_type):
exp_data = {
'sparseness': sparseness,
'index': index,
'extend': extend_near_num,
'percentage': percentage
}
matrix = create_sparse_matrix(
load_training_data(sparseness,
index,
extend_near_num,
matrix_type=matrix_type))
U, Sigma, VT = la.svd(matrix)
K = sigma_pct(Sigma, percentage)
U1 = U[:, :K]
VT1 = VT[:K, :]
Sigma1 = np.eye(K) * Sigma[:K]
R = np.matmul(np.matmul(U1, Sigma1), VT1)
mae, rmse = evaluate(sparseness, index, R, matrix_type=matrix_type)
exp_data["mae"] = float(mae)
exp_data["rmse"] = float(rmse)
exp_data['datetime'] = datetime.now()
print(exp_data)
auto_insert_database(database_remote_config, exp_data,
f'svd_{matrix_type}')
def uppc_experiment(**kwargs):
exp_data = {}
exp_data.update(kwargs)
sparseness = kwargs['sparseness']
index = kwargs['index']
extend = kwargs['extend']
matrix_type = kwargs['matrix_type']
matrix = create_sparse_matrix(
load_training_data(sparseness, index, extend, matrix_type=matrix_type))
test_userId, test_itemId, test_rating = \
load_data(csv_file_path(sparseness, index, training_set=False, matrix_type=matrix_type))
R = np.corrcoef(matrix)
R = np.nan_to_num(R, nan=-1)
def predict(user, item):
i = matrix[:, item]
i[i >= 0] = 1
return np.matmul(R[user], matrix[:, item]) / np.sum(
np.abs(np.matmul(R[user], i)))
test_predict = np.array(
list(map(lambda u, i: predict(u, i), test_userId, test_itemId)))
mae = np.mean(np.abs(test_predict - test_rating))
rmse = np.sqrt(np.mean(np.square(test_predict - test_rating)))
exp_data["mae"] = float(mae)
exp_data["rmse"] = float(rmse)
exp_data['datetime'] = datetime.now()
print(exp_data)
auto_insert_database(database_remote_config, exp_data,
f'uppc_{matrix_type}')
def experiment(**kwargs):
sparseness = kwargs['sparseness']
index = kwargs['index']
epochs = kwargs['epochs']
batch_size = kwargs['batch_size']
layers = kwargs['layers']
reg_layers = kwargs['reg_layers']
fake_layers = kwargs['fake_layers']
fake_reg_layers = kwargs['fake_reg_layers']
last_activation = kwargs['last_activation']
fake_last_activation = kwargs['fake_last_activation']
learning_rate = kwargs['learning_rate']
extend_near_num = kwargs['extend_near_num']
learner = kwargs['learner']
exp_data = {
'sparseness': sparseness,
'index': index,
'epochs': epochs,
'batch_size': batch_size,
'layers': layers,
'reg_layers': reg_layers,
'fake_layers': fake_layers,
'fake_reg_layers': fake_reg_layers,
'last_activation': last_activation,
'fake_last_activation': fake_last_activation,
'learning_rate': learning_rate,
'extend_near_num': extend_near_num,
'learner': learner
}
optimizer = {'adagrad': Adagrad(lr=learning_rate),
'rmsprop': RMSprop(lr=learning_rate),
'adam': Adam(lr=learning_rate),
'sgd': SGD(lr=learning_rate)}[learner]
dataset_name = 'sparseness%s_%s' % (sparseness, index)
model_out_file = '%s_exMLP_%s_%s_%s.h5' % (dataset_name, layers, fake_layers, datetime.now())
userId, itemId, rating = load_data(csv_file_path(sparseness, index))
result = calculate_distance(csv_file_path(sparseness, index))
distance = convert_distance_result(result)
fake_user_id, userId, itemId, rating = extend_array(extend_near_num, distance, userId, itemId, rating)
test_userId, test_itemId, test_rating = load_data(csv_file_path(sparseness, index, training_set=False))
# early_stop = keras.callbacks.EarlyStopping(monitor='mean_absolute_error', min_delta=0.0002, patience=10)
model = custom_model.extend_mlp_model.get_model(num_users=user_num, num_items=ws_num,
layers=layers, reg_layers=reg_layers,
fake_layers=fake_layers, fake_reg_layers=fake_reg_layers,
last_activation=last_activation,
fake_last_activation=fake_last_activation)
model.compile(optimizer=optimizer)
model.fit(x=[fake_user_id, userId, itemId, rating],
y=None,
batch_size=batch_size, epochs=epochs,
verbose=1,
shuffle=False)
mkdir('./Trained')
model.save('./Trained/{}'.format(model_out_file))
# prediction, fake_prediction = model.predict([np.zeros(len(test_userId)), test_userId, test_itemId])
# mae = np.mean(np.abs(prediction - test_rating))
# rmse = np.sqrt(np.mean(np.square(prediction - test_rating)))
_, _, mae, rmse = model.evaluate([np.zeros(len(test_rating)), test_userId, test_itemId, test_rating], steps=1)
# print('loss: ', loss)
# print('mae: ', mae)
# print('rmse', np.sqrt(mse))
exp_data['model'] = model_out_file
exp_data['mae'] = float(mae)
exp_data['rmse'] = float(rmse)
exp_data['datetime'] = datetime.now()
exp_data['last_activation'] = last_activation
print(exp_data)
auto_insert_database(database_config, exp_data, 'exmlp_rt')