How can I test individual layers in TensorFlow?

Question

I have constructed a 7 layer convolutional network, based off of the DEEP MNIST Expert tutorial. I have added two more convolutional layers.

Everything runs well, but I would like to attempt to input 1024 x 10 arrays directly into the fully connected layer, and circumvent the convolutional layers.

Is there any way to do this without rebuilding the entire network?

Answer 1

Between the convolutional layers and the fully connected layer, create a place holder for the input to the fully connected layer: input_to_fc = tf.placeholder_with_default(previous_layer, shape=(None, 1024*10)) . You can bypass the convolutional layers by feeding the input directly to the input_to_fc tensor.

Example:

...
conv = tf.layers.conv2d(...)
flatten = tf.layers.flatten(...)
input_to_fc = tf.placeholder_with_default(flatten, shape=OUTPUT_SHAPE_OF_PREVIOUS_LAYER))
fc = tf.layers.dense(input_to_fc, ...)