I am playing with a ANN which is part of Udacity DeepLearning course.
I have an assignment which involves introducing generalization to the network with one hidden ReLU layer using L2 loss. I wonder how to properly introduce it so that ALL weights are penalized, not only weights of the output layer.
Code for network without generalization is at the bottom of the post (code to actually run the training is out of the scope of the question).
Obvious way of introducing the L2 is to replace the loss calculation with something like this (if beta is 0.01):
loss = tf.reduce_mean( tf.nn.softmax_cross_entropy_with_logits(out_layer, tf_train_labels) + 0.01*tf.nn.l2_loss(out_weights))
But in such case it will take into account values of output layer's weights. I am not sure, how do we properly penalize the weights which come INTO the hidden ReLU layer. Is it needed at all or introducing penalization of output layer will somehow keep the hidden weights in check also?
#some importing
from __future__ import print_function
import numpy as np
import tensorflow as tf
from six.moves import cPickle as pickle
from six.moves import range
#loading data
pickle_file = '/home/maxkhk/Documents/Udacity/DeepLearningCourse/SourceCode/tensorflow/examples/udacity/notMNIST.pickle'
with open(pickle_file, 'rb') as f:
save = pickle.load(f)
train_dataset = save['train_dataset']
train_labels = save['train_labels']
valid_dataset = save['valid_dataset']
valid_labels = save['valid_labels']
test_dataset = save['test_dataset']
test_labels = save['test_labels']
del save # hint to help gc free up memory
print('Training set', train_dataset.shape, train_labels.shape)
print('Validation set', valid_dataset.shape, valid_labels.shape)
print('Test set', test_dataset.shape, test_labels.shape)
#prepare data to have right format for tensorflow
#i.e. data is flat matrix, labels are onehot
image_size = 28
num_labels = 10
def reformat(dataset, labels):
dataset = dataset.reshape((-1, image_size * image_size)).astype(np.float32)
# Map 0 to [1.0, 0.0, 0.0 ...], 1 to [0.0, 1.0, 0.0 ...]
labels = (np.arange(num_labels) == labels[:,None]).astype(np.float32)
return dataset, labels
train_dataset, train_labels = reformat(train_dataset, train_labels)
valid_dataset, valid_labels = reformat(valid_dataset, valid_labels)
test_dataset, test_labels = reformat(test_dataset, test_labels)
print('Training set', train_dataset.shape, train_labels.shape)
print('Validation set', valid_dataset.shape, valid_labels.shape)
print('Test set', test_dataset.shape, test_labels.shape)
#now is the interesting part - we are building a network with
#one hidden ReLU layer and out usual output linear layer
#we are going to use SGD so here is our size of batch
batch_size = 128
#building tensorflow graph
graph = tf.Graph()
with graph.as_default():
# Input data. For the training data, we use a placeholder that will be fed
# at run time with a training minibatch.
tf_train_dataset = tf.placeholder(tf.float32,
shape=(batch_size, image_size * image_size))
tf_train_labels = tf.placeholder(tf.float32, shape=(batch_size, num_labels))
tf_valid_dataset = tf.constant(valid_dataset)
tf_test_dataset = tf.constant(test_dataset)
#now let's build our new hidden layer
#that's how many hidden neurons we want
num_hidden_neurons = 1024
#its weights
hidden_weights = tf.Variable(
tf.truncated_normal([image_size * image_size, num_hidden_neurons]))
hidden_biases = tf.Variable(tf.zeros([num_hidden_neurons]))
#now the layer itself. It multiplies data by weights, adds biases
#and takes ReLU over result
hidden_layer = tf.nn.relu(tf.matmul(tf_train_dataset, hidden_weights) + hidden_biases)
#time to go for output linear layer
#out weights connect hidden neurons to output labels
#biases are added to output labels
out_weights = tf.Variable(
tf.truncated_normal([num_hidden_neurons, num_labels]))
out_biases = tf.Variable(tf.zeros([num_labels]))
#compute output
out_layer = tf.matmul(hidden_layer,out_weights) + out_biases
#our real output is a softmax of prior result
#and we also compute its cross-entropy to get our loss
loss = tf.reduce_mean( tf.nn.softmax_cross_entropy_with_logits(out_layer, tf_train_labels))
#now we just minimize this loss to actually train the network
optimizer = tf.train.GradientDescentOptimizer(0.5).minimize(loss)
#nice, now let's calculate the predictions on each dataset for evaluating the
#performance so far
# Predictions for the training, validation, and test data.
train_prediction = tf.nn.softmax(out_layer)
valid_relu = tf.nn.relu( tf.matmul(tf_valid_dataset, hidden_weights) + hidden_biases)
valid_prediction = tf.nn.softmax( tf.matmul(valid_relu, out_weights) + out_biases)
test_relu = tf.nn.relu( tf.matmul( tf_test_dataset, hidden_weights) + hidden_biases)
test_prediction = tf.nn.softmax(tf.matmul(test_relu, out_weights) + out_biases)
hidden_weights, hidden_biases, out_weights, and out_biases are all the model parameters that you are creating. You can add L2 regularization to ALL these parameters as follows :
loss = (tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(
logits=out_layer, labels=tf_train_labels)) +
0.01*tf.nn.l2_loss(hidden_weights) +
0.01*tf.nn.l2_loss(hidden_biases) +
0.01*tf.nn.l2_loss(out_weights) +
0.01*tf.nn.l2_loss(out_biases))
A shorter and scalable way of doing this would be ;
vars = tf.trainable_variables()
lossL2 = tf.add_n([ tf.nn.l2_loss(v) for v in vars ]) * 0.001
This basically sums the l2_loss of all your trainable variables. You could also make a dictionary where you specify only the variables you want to add to your cost and use the second line above. Then you can add lossL2 with your softmax cross entropy value in order to calculate your total loss.
Edit : As mentioned by Piotr Dabkowski, the code above will also regularise biases. This can be avoided by adding an if statement in the second line ;
lossL2 = tf.add_n([ tf.nn.l2_loss(v) for v in vars
if 'bias' not in v.name ]) * 0.001
This can be used to exclude other variables.
In fact, we usually do not regularize bias terms (intercepts). So, I go for:
loss = (tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(
logits=out_layer, labels=tf_train_labels)) +
0.01*tf.nn.l2_loss(hidden_weights) +
0.01*tf.nn.l2_loss(out_weights))
By penalizing the intercept term, as the intercept is added to y values, it will result in changing the y values, adding a constant c to the intercepts. Having it or not will not change the results but takes some computations
来源：https://stackoverflow.com/questions/38286717/tensorflow-regularization-with-l2-loss-how-to-apply-to-all-weights-not-just