def test_mean_squared_error():
from linear_regression import mean_squared_error
outputs = np.float32([[1, 2, 3, 4]]).T
targets = np.float32([[1.1, 2.2, 3.3, 4.4]]).T
actual = mean_squared_error(outputs, targets)
desired = 0.07500000000000007
np.testing.assert_allclose(actual, desired, rtol=1e-3, atol=1e-3)
with open("../validation_set.json", "r") as rf_validation_set:
data = json.load(rf_validation_set)
X_validation_set_Adv, Y_validation_set_Adv, str_output = pf.generate_wordfeature_and_output(wordcount, data, False, 0, False, 0)
#測試Closed form的性能
start_time = time.time()
W_closed, str_to_write = lr.least_squares_estimate_linear_regression_alg(X_validation_set_Adv, Y_validation_set_Adv)
closed_running_time = time.time() - start_time
est_Y = np.dot(X_training_set_Adv, W_closed)
trn_mse_closed, diff_AB, Sigma_Square_of_diff_AB = lr.mean_squared_error(est_Y, Y_training_set_Adv)
est_Y = np.dot(X_validation_set_Adv, W_closed)
val_mse_closed, diff_AB, Sigma_Square_of_diff_AB = lr.mean_squared_error(est_Y, Y_validation_set_Adv)
#測試Gradient Descent的性能
list_T_Robbins_Monroe_pow = []
list_trn_mse = []
list_val_mse = []
list_elapsed_time = []
list_W_MSE = []
epsilon_power = -6
eta_power = 0
for T_Robbins_Monroe_pow in range(1, 8, 1):
str_for_statistics += "least_squares_estimate_linear_regression_alg:" + "\n\n"
running_stat_item['desc'] = "least_squares_estimate_linear_regression_alg"
str_for_statistics += "W_cf = \n" + str(W_cf) + "\n\n"
running_stat_item['W'] = str(W_cf)
str_for_statistics += "alg runtime = " + str(elapsed_time) + "\n"
running_stat_item['alg_runtime'] = elapsed_time
start_time = time.time()
est_Y = np.dot(X_training_set_Adv, W_cf)
tmp_mse, diff_AB, Sigma_Square_of_diff_AB = lr.mean_squared_error(
est_Y, Y_training_set_Adv)
#print("est_Y = \n", est_Y)
#print("tmp_mse =", tmp_mse)
elapsed_time = time.time() - start_time
str_to_write = "tmp_mse for training set = " + str(tmp_mse) + "\n\n\n\n"
str_to_write += "mean_squared_error elapsed_time = " + str(
elapsed_time) + "\n"
Sigma_Square_of_diff_AB = np.sort(Sigma_Square_of_diff_AB,
axis=None,
kind='mergesort')
str_to_write += "Sigma_Square_of_diff_AB = \n" + str(
Sigma_Square_of_diff_AB) + "\n" + str(
fout_t2r.write(str_to_write)
print(str_to_write)
str_to_write = "lr.least_squares_estimate_linear_regression_alg W = \n" + str(
W) + "\n"
elapsed_time = time.time() - start_time
str_to_write += "elapsed_time = " + str(elapsed_time) + "\n"
fout_t2r.write(str_to_write)
print(str_to_write)
#print("elapsed_time = ", elapsed_time)
est_Y = np.dot(X_validation_set, W)
tmp_mse = lr.mean_squared_error(est_Y, Y_validation_set)
#print("est_Y = \n", est_Y)
#print("tmp_mse =", tmp_mse)
str_to_write = "est_Y = \n" + str(est_Y) + "\n"
str_to_write += "tmp_mse = " + str(tmp_mse) + "\n"
fout_t2r.write(str_to_write)
print(str_to_write)
# -2 ~ -8
epsilon_power = -2
# 1 ~ -3
beta_power = 1