How can I access weights from a layer created with tf.contrib.layers.fully_connected?

Question

I'm using tf.contrib.layers.fully_connected to create a layer in the following code.

library(tensorflow)

x <- tf$placeholder(tf$float32, shape(NULL, 784L))
logits <- tf$contrib$layers$fully_connected(x, 10L)
y <- tf$nn$softmax(logits)

How can I access the weights as I would in the following block of code with sess$run(W) ?

x <- tf$placeholder(tf$float32, shape(NULL, 784L))
W <- tf$Variable(tf$zeros(shape(784L, 10L)))
b <- tf$Variable(tf$zeros(shape(10L)))
y <- tf$nn$softmax(tf$matmul(x, W) + b)

Note: I'm using TensorFlow for R but this should be the same as TensorFlow for Python by changing $ for . .

Answer 1

You can get the list of all global variables using tf$global_variables() . Not an ideal solution (because it retrieves a list of unnamed variables), but it should get you what you need. Reproducible example below

library(tensorflow)

datasets <- tf$contrib$learn$datasets
mnist <- datasets$mnist$read_data_sets("MNIST-data", one_hot = TRUE)

x <- tf$placeholder(tf$float32, shape(NULL, 784L))
logits <- tf$contrib$layers$fully_connected(x, 10L)

y <- tf$nn$softmax(logits)
y_ <- tf$placeholder(tf$float32, shape(NULL,10L))

cross_entropy <- tf$reduce_mean(-tf$reduce_sum(y_ * tf$log(y), reduction_indices=1L))
train_step <- tf$train$GradientDescentOptimizer(0.5)$minimize(cross_entropy)

sess <- tf$Session()
sess$run(tf$global_variables_initializer())

for (i in 1:1000) {
  batches <- mnist$train$next_batch(100L)
  batch_xs <- batches[[1]]
  batch_ys <- batches[[2]]
  sess$run(train_step,
           feed_dict = dict(x = batch_xs, y_ = batch_ys))
}

lst.variables <- sess$run(tf$global_variables())
str(lst.variables)

Answer 2

You pass the name of the tensor to the run function. You should inspect the graph to see the name of the tensor added to the graph from the function.