def runner_with_results_output(y_bundle):
test_losses = []
test_accuracies = []
for i in range(batch_size):
test_losses.append(3.5 - 1.6 * sigmoid(i / 10))
test_accuracies.append(0.5 + 0.4 * sigmoid(i / 10))
y = YaxisBundle(test_losses, "loss", "b")
y_bundle.append(y)
y = YaxisBundle(test_accuracies, "accuracy", "g")
y_bundle.append(y)
return y_bundle
def run_with_config(config): #, X_train, y_train, X_test, y_test):
tf.reset_default_graph() # To enable to run multiple things in a loop
config.print_config()
if config.matplot_lib_enabled:
# To keep track of training's performance
test_losses = []
test_accuracies = []
indep_test_axis = []
config.W = {
'hidden':
tf.Variable(tf.random_normal([config.n_inputs, config.n_hidden])),
'output':
tf.Variable(tf.random_normal([config.n_hidden, config.n_classes]))
}
config.biases = {
'hidden': tf.Variable(tf.random_normal([config.n_hidden], mean=1.0)),
'output': tf.Variable(tf.random_normal([config.n_classes]))
}
#-----------------------------------
# Define parameters for model
#-----------------------------------
print(
"Some useful info to get an insight on dataset's shape and normalisation:"
)
print(
"features shape, labels shape, each features mean, each features standard deviation"
)
print(X_test.shape, y_test.shape, np.mean(X_test), np.std(X_test))
print("the dataset is therefore properly normalised, as expected.")
# ------------------------------------------------------
# step3: Let's get serious and build the neural network
# ------------------------------------------------------
X = tf.placeholder(tf.float32, [None, config.n_steps, config.n_inputs])
Y = tf.placeholder(tf.float32, [None, config.n_classes])
pred_Y = LSTM_Network(X, config)
print "Unregularised variables:"
for unreg in [
tf_var.name for tf_var in tf.trainable_variables()
if ("noreg" in tf_var.name or "Bias" in tf_var.name)
]:
print unreg
# Loss,optimizer,evaluation
l2 = config.lambda_loss_amount * \
sum(tf.nn.l2_loss(tf_var) for tf_var in tf.trainable_variables())
# Softmax loss and L2
cost = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(logits=pred_Y, labels=Y)) + l2
optimizer = tf.train.AdamOptimizer(
learning_rate=config.learning_rate).minimize(cost)
correct_pred = tf.equal(tf.argmax(pred_Y, 1), tf.argmax(Y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, dtype=tf.float32))
# ------------------------------------------------------
# step3.5 : Tensorboard stuff here
# ------------------------------------------------------
if config.tensor_board_logging_enabled:
tf.summary.scalar("loss", cost)
tf.summary.scalar("accuracy", accuracy)
merged_summary_op = tf.summary.merge_all()
# --------------------------------------------
# step4: Hooray, now train the neural network
# --------------------------------------------
# Note that log_device_placement can be turned ON but will cause console spam.
sess = tf.InteractiveSession(config=tf.ConfigProto(
log_device_placement=False))
tf.global_variables_initializer().run()
if config.tensor_board_logging_enabled:
# op to write logs to Tensorboard
summary_writer = tf.summary.FileWriter(config.logs_path,
graph=tf.get_default_graph())
best_accuracy = 0.0
# Start training for each batch and loop epochs
for i in range(config.training_epochs):
for start, end in zip(
range(0, config.train_count, config.batch_size),
range(config.batch_size, config.train_count + 1,
config.batch_size)):
if config.tensor_board_logging_enabled:
_, summary = sess.run([optimizer, merged_summary_op],
feed_dict={
X: X_train[start:end],
Y: y_train[start:end]
})
else:
sess.run(optimizer,
feed_dict={
X: X_train[start:end],
Y: y_train[start:end]
})
if config.tensor_board_logging_enabled:
# Write logs at every iteration
summary_writer.add_summary(summary, i)
# Test completely at every epoch: calculate accuracy
pred_out, accuracy_out, loss_out = sess.run([pred_Y, accuracy, cost],
feed_dict={
X: X_test,
Y: y_test
})
if config.matplot_lib_enabled:
indep_test_axis.append(i)
test_losses.append(loss_out)
test_accuracies.append(accuracy_out)
print("traing iter: {},".format(i) + \
" test accuracy : {},".format(accuracy_out) + \
" loss : {}".format(loss_out))
best_accuracy = max(best_accuracy, accuracy_out)
print("")
print("final test accuracy: {}".format(accuracy_out))
print("best epoch's test accuracy: {}".format(best_accuracy))
print("")
if config.tensor_board_logging_enabled:
print("Run the command line:\n")
print(config.tensorboard_cmd)
print("\nThen open http://0.0.0.0:6006/ into your web browser")
print(config.model_desc_attched_string)
if config.matplot_lib_enabled:
#for i in range(config.batch_size):
# indep_test_axis.append(i)
#indep_test_axis = [i for i in range(config.batch_size)]
#indep_test_axis = np.array(indep_test_axis)
#p = PlotUtil("title", indep_test_axis, "x_label", "y_label")
y_bundle = []
test_losses = np.array(test_losses)
test_accuracies = np.array(test_accuracies)
y = YaxisBundle(np.array(test_losses), "loss", "b")
y_bundle.append(y)
y = YaxisBundle(np.array(test_accuracies), "accuracy", "g")
y_bundle.append(y)
#p.show_plot(y_bundle)
if config.matplot_lib_for_single_ybundle:
if config.matplot_lib_for_accuracy:
return y_bundle[1]
else:
return y_bundle[0]
return y_bundle