def run_task3(train_samples, test_samples, learning_rate, epochs, print_option='default'):
spot_0 = time.time()
train_X, train_y = create_samples_mul(train_samples)
test_X, test_y = create_samples_mul(test_samples)
train_X, train_y = train_X.reshape(-1,1,1,2), train_y.reshape(-1,1,1,1)
test_X, test_y = test_X.reshape(-1,1,1,2), test_y.reshape(-1,1,1,1)
# make model
spot_1 = time.time()
model = nn.Network(
layers = [
nn.Layer(INPUT_DIM=(1,1,2), kernel=(1,1,2,3), stride=(1,1), OUTPUT_DIM=(1,1,3)),
nn.Layer(INPUT_DIM=(1,1,3), kernel=(1,1,3,1), stride=(1,1), OUTPUT_DIM=(1,1,1)),
]
)
if print_option != 'nothing':
print('train_samples : ', train_samples, train_X.shape)
print('test_samples : ', test_samples, train_y.shape)
print('epochs : ', epochs)
print('learning_rate : ', learning_rate)
model.print_layers();
spot_2 = time.time()
for epoch in range(epochs):
model.train(train_X, train_y, learning_rate)
if print_option == 'progress':
print('\nepoch #' + str(epoch+1))
print('w :' + str(model.layers[0].w.reshape((1,2))))
print('b : ' + str(model.layers[0].b))
spot_3 = time.time()
test_pred = model.predict(test_X)
spot_4 = time.time()
train_acc = 100 * np.mean(model.predict(train_X).astype(bool) == train_y.astype(bool))
test_acc = 100 * np.mean(test_pred.astype(bool) == test_y.astype(bool))
return {'w0': model.layers[0].w.reshape((3,2)),
'b0': model.layers[0].b.reshape((3,1)),
'w1': model.layers[1].w.reshape((1,3)),
'b1': model.layers[1].b,
'train_loss': model.loss(train_X, train_y),
'test_loss': model.loss(test_X, test_y),
'train_acc': train_acc,
'test_acc': test_acc,
'time_gen_data' : spot_1-spot_0,
'time_gen_network' : spot_2-spot_1,
'time_train' : spot_3-spot_2,
'time_predict' : spot_4-spot_3,
}
def run_practice_2(train_samples,
test_samples,
learning_rate,
epochs,
print_option='default'):
train_X, train_y = create_samples_sum(train_samples)
test_X, test_y = create_samples_sum(test_samples)
train_X, train_y = train_X.reshape(-1, 1, 1,
2), train_y.reshape(-1, 1, 1, 1)
test_X, test_y = test_X.reshape(-1, 1, 1, 2), test_y.reshape(-1, 1, 1, 1)
# make model
model = nn.Network(layers=[
nn.Layer(INPUT_DIM=(1, 1, 2),
kernel=(1, 1, 2, 1),
stride=(1, 1),
OUTPUT_DIM=(1, 1, 1)),
])
if print_option != 'nothing':
print('train_samples : ', train_samples, train_X.shape)
print('test_samples : ', test_samples, train_y.shape)
print('epochs : ', epochs)
print('learning_rate : ', learning_rate)
model.print_layers()
for epoch in range(epochs):
model.train(train_X, train_y, learning_rate)
if print_option == 'progress':
print('\nepoch #' + str(epoch + 1))
print('w :' + str(model.layers[0].w.reshape((1, 2))))
print('b : ' + str(model.layers[0].b))
return {
'w':
model.layers[0].w.reshape((1, 2)),
'b':
model.layers[0].b,
'train_loss':
model.loss(train_X, train_y),
'test_loss':
model.loss(test_X, test_y),
'train_acc':
100 *
np.mean(model.predict(train_X).astype(bool) == train_y.astype(bool)),
'test_acc':
100 *
np.mean(model.predict(test_X).astype(bool) == test_y.astype(bool))
}