c = get_label_data("t10k-labels-idx1-ubyte")
y = get_feature_data("t10k-images-idx3-ubyte")
# cTrain = get_label_data("train-labels-idx1-ubyte")[:,:10000]
# yTrain = get_feature_data("train-images-idx3-ubyte")[:10000, :]
cTrain = get_label_data("train-labels-idx1-ubyte")
yTrain = get_feature_data("train-images-idx3-ubyte")
y = np.matrix(normalize(y, axis=1))
yTrain = np.matrix(normalize(yTrain, axis=1))
# print(yTrain.shape)
# print(cTrain.shape)
w_init = get_weight_matrix(yTrain, cTrain)
score1 = calculate_accuracy(y, w_init, c)
print("initial accuracy", score1)
# Testing CGLS
# hessian_handle = lambda x: hessian_sub(yTrain.T, w_init.T, x)
# D = cgls_wiki(
# hessian_handle,
# -1 * softmax_gradient(cTrain, yTrain.T, w_init.T),
# k=200,
# x_0=w_init.T
# )
# diff_matrix = -1 * softmax_gradient(cTrain, yTrain.T, w_init.T) - hessian_sub(yTrain.T,w_init.T, D)
# print("norm of diff_matrix", np.linalg.norm(diff_matrix))
return 1 / (1 + np.exp(-values))
if __name__ == '__main__':
data_formatted, data_without_class, data_class_only, _ = func.read_csv_and_prep(
)
# Initialise coefficients (weights) to 0
coefficients = np.zeros(data_formatted.shape[1])
predictions = None
for count in range(EPOCH_COUNT):
if count > 0 and count % 10000 == 0:
print("Count: ", count)
predictions = logistic_function(data_formatted, coefficients)
# Calculating deviation between real and predicted
error = data_class_only.T - predictions
# The gradient function is to be reduced, as we approach 0 we approach the least error (as the gradient tends
# towards 0, our function shows that our results are becoming more accurate)
gradient = np.dot(data_formatted.T, error)
# Adjusting coefficients for more accurate result
coefficients += gradient * LEARNING_RATE
print("Coefficient values {}".format(coefficients))
print("Accuracy: {}".format(
func.calculate_accuracy(predictions, data_class_only)))
for item in N_trees_list:
start_train_time = time.time()
forest = random_forest_algorithm(train_df,
n_trees=item,
n_bootstrap=100,
n_features=100,
dt_max_depth=10)
train_time = time.time() - start_train_time
start_test_time = time.time()
predictions = random_forest_predictions(test_df, forest)
test_time = time.time() - start_test_time
accuracy = calculate_accuracy(predictions, test_df.Label)
accuracy_list.append(accuracy)
train_time_list.append(train_time)
test_time_list.append(test_time)
title_name = str(
'accuracy_k120_axis-aligned_varytrees2_bootstrap100_features100_depth10.png'
)
fig = plt.figure()
ax = fig.add_subplot(111)
lns1 = ax.plot(N_trees_list, train_time_list, '-r', label='train_time')
lns2 = ax.plot(N_trees_list, test_time_list, '-g', label='test_time')
ax2 = ax.twinx()