def run(constant_overwrites):
x_train, y_train, x_val, y_val, x_test, y_test = load_mnist_dataset(
flatten=True)
plt.figure(figsize=[6, 6])
for i in range(4):
plt.subplot(2, 2, i + 1)
plt.title('Label: %i' % y_train[i])
plt.imshow(x_train[i].reshape([28, 28]), cmap='gray')
constants = merge_dict(get_constants(), constant_overwrites)
n_classes = constants['n_classes']
data = {
'X_train': x_train,
'y_train': one_hot_encode(y_train, n_classes),
'X_val': x_val,
'y_val': one_hot_encode(y_val, n_classes),
'X_test': x_test,
'y_test': one_hot_encode(y_test, n_classes)
}
input_x, input_y = get_inputs(constants)
parameters = get_parameters(constants)
optimizer, loss_op, model, y_ = model_builder(network_builder, input_x,
input_y, parameters,
constants)
train(data,
constants, (input_x, input_y),
optimizer,
loss_op,
model,
y_,
minibatch=True)
def load_data_and_labels(df, n_classes):
x = [clean_str(sent) for sent in df.x]
y = one_hot_encode(df.y.values, n_classes)
return x, y
def run(constant_overwrites):
constants = merge_dict(get_constants(), constant_overwrites)
x_train, y_train, x_val, y_val, x_test, y_test = load_mnist_dataset(
flatten=True)
if constants['retrain']:
plt.figure(figsize=[6, 6])
for i in range(4):
plt.subplot(2, 2, i + 1)
plt.title('Label: %i' % y_train[i])
plt.imshow(x_train[i].reshape([28, 28]), cmap='gray')
plt.show()
n_classes = constants['n_classes']
data = {
'X_train': x_train,
'y_train': one_hot_encode(y_train, n_classes),
'X_val': x_val,
'y_val': one_hot_encode(y_val, n_classes),
'X_test': x_test,
'y_test': one_hot_encode(y_test, n_classes)
}
input_x, input_y = get_inputs(constants)
parameters = get_parameters(constants)
optimizer, loss_op, predict_op, model, y_ = \
model_builder(network_builder, input_x, input_y, parameters, constants)
saver = tf.train.Saver()
train(data,
constants, (input_x, input_y),
optimizer,
loss_op,
model,
y_,
minibatch=True,
saver=saver)
app_name = constants['app_name']
model_name = constants['model_name']
version = constants['version']
model_dir = 'data'
predict_op_name = 'predict_op:0'
input_name = 'input_X:0'
sess = tf.Session()
saver = tf.train.import_meta_graph('{}/{}.meta'.format(
model_dir, model_name))
saver.restore(sess, tf.train.latest_checkpoint(model_dir))
print('Testing...:')
sample = x_test[random.randint(0, len(x_test))]
plt.imshow(sample.reshape([28, 28]), cmap='gray')
plt.show()
pred = sess.run(predict_op_name, feed_dict={input_name: [sample]})
print('prediction:', pred[0])
if constants['serve']:
print('Serving model.')
serve(app_name=app_name,
model_name=model_name,
model_dir=model_dir,
version=version,
predict_op_name=predict_op_name,
input_name=input_name,
input_type='floats',
default_output='-1.0')