Creates loss tensor for resnet model.

fully connected layer

Pads the input along the spatial dimensions independently of input size. Args: inputs: A tensor of size [batch, channels, height_in, width_in]

Helper method for printing ops. Converts Tensor/Operation op to op.name, rest to str(op).

Takes A0, applies n operations to it, returns An.

Creates dictionary {node: {child1, child2, ..},..} for current TensorFlow graph. Args: g: tf.Graph to use or None, in which case take default

Creates loss tensor for resnet model.

Generator for ImageNet ResNet v2 models. Args: block_fn: The block to use within the model, either `building_block` or `bottleneck_block`

numerically stable log_softmax implementation that prevents overflow

Creates chain of length length. First node is Variable, rest are tanh. Returns nodes. Note, if length is 1, there are no non-linearities in the gr

Takes the MNIST inputs and mode and outputs a tensor of logits.

a network in network layer (1x1 CONV)

Return dictionary of peak memory usage (bytes) for each device. {"cpu:0": 20441, ...

Return parts of the graph needed to compute targets.

Remove ops which either require initialization (Variable) or update values. This ensure that linearizing the whole graph doesn't break initializatio

Make sure articulation points are found correctly in resnet.

Runs regular chain gradient, makes sure memory usage makes sense.

Take chain of length 5, save 2 nodes, make sure 2 units of RAM is saved.

Take chain of length 5, save first node.

Take chain of length 5, save last node. This saved no memory, and is and edge case that should raise exception by rewriter.

Take chain of length 5, save first node.

Runs regular chain gradient, makes sure memory usage makes sense.

Runs regular chain gradient, makes sure memory usage makes sense.

sanity check on TF resnet structure.

Creates op->children dictionary from list of TensorFlow ops.

Generator for ImageNet ResNet v2 models. Args: block_fn: The block to use within the model, either `building_block` or `bottleneck_block`

Returns a tf.data.Dataset yielding (image, label) pairs for training.

- data_dir is location where the files are stored - subset is train|test - batch_size is int, of #examples to load at once

- data_dir is location where to store files - subset is train|test - batch_size is int, of #examples to load at once

n is the number of examples to fetch

n is the number of examples to fetch

Adam optimizer

Bottleneck block variant for residual networks with BN before convolutions. Args: inputs: A tensor of size [batch, channels, height_in, width_i

Standard building block for residual networks with BN before convolutions. Args: inputs: A tensor of size [batch, channels, height_in, width_in

Decorator to capture ops created in the block. with capture_ops() as ops: # create some ops print(ops) # => prints ops created.

Decorator to capture ops created in the block. with capture_ops() as ops: # create some ops print(ops) # => prints ops created.

Generator for CIFAR-10 ResNet v2 models. Args: resnet_size: A single integer for the size of the ResNet model. num_classes: The number of p