View source on GitHub |
Ops for building neural network layers, regularizers, summaries, etc.
Modules
feature_column
module: This API defines FeatureColumn abstraction.
summaries
module: Utility functions for summary creation.
Classes
class GDN
: Generalized divisive normalization layer.
class RevBlock
: Block of reversible layers. See rev_block.
Functions
apply_regularization(...)
: Returns the summed penalty by applying regularizer
to the weights_list
.
avg_pool2d(...)
: Adds a 2D average pooling op.
avg_pool3d(...)
: Adds a 3D average pooling op.
batch_norm(...)
: Adds a Batch Normalization layer from http://arxiv.org/abs/1502.03167
bias_add(...)
: Adds a bias to the inputs.
bow_encoder(...)
: Maps a sequence of symbols to a vector per example by averaging embeddings.
bucketized_column(...)
: Creates a _BucketizedColumn for discretizing dense input.
check_feature_columns(...)
: Checks the validity of the set of FeatureColumns.
conv1d(...)
: Adds an N-D convolution followed by an optional batch_norm layer.
conv2d(...)
: Adds an N-D convolution followed by an optional batch_norm layer.
conv2d_in_plane(...)
: Performs the same in-plane convolution to each channel independently.
conv2d_transpose(...)
: Adds a convolution2d_transpose with an optional batch normalization layer.
conv3d(...)
: Adds an N-D convolution followed by an optional batch_norm layer.
conv3d_transpose(...)
: Adds a convolution3d_transpose with an optional batch normalization layer.
convolution(...)
: Adds an N-D convolution followed by an optional batch_norm layer.
convolution1d(...)
: Adds an N-D convolution followed by an optional batch_norm layer.
convolution2d(...)
: Adds an N-D convolution followed by an optional batch_norm layer.
convolution2d_in_plane(...)
: Performs the same in-plane convolution to each channel independently.
convolution2d_transpose(...)
: Adds a convolution2d_transpose with an optional batch normalization layer.
convolution3d(...)
: Adds an N-D convolution followed by an optional batch_norm layer.
convolution3d_transpose(...)
: Adds a convolution3d_transpose with an optional batch normalization layer.
create_feature_spec_for_parsing(...)
: Helper that prepares features config from input feature_columns.
crossed_column(...)
: Creates a _CrossedColumn for performing feature crosses.
dense_to_sparse(...)
: Converts a dense tensor into a sparse tensor.
dropout(...)
: Returns a dropout op applied to the input.
elu(...)
: partial(func, *args, **keywords) - new function with partial application
embed_sequence(...)
: Maps a sequence of symbols to a sequence of embeddings.
embedding_column(...)
: Creates an _EmbeddingColumn
for feeding sparse data into a DNN.
embedding_lookup_unique(...)
: Version of embedding_lookup that avoids duplicate lookups.
flatten(...)
: Flattens the input while maintaining the batch_size.
fully_connected(...)
: Adds a fully connected layer.
gdn(...)
: Functional interface for GDN layer.
group_norm(...)
: Functional interface for the group normalization layer.
images_to_sequence(...)
: Convert a batch of images into a batch of sequences.
infer_real_valued_columns(...)
input_from_feature_columns(...)
: A tf.contrib.layers style input layer builder based on FeatureColumns.
instance_norm(...)
: Functional interface for the instance normalization layer.
joint_weighted_sum_from_feature_columns(...)
: A restricted linear prediction builder based on FeatureColumns.
l1_l2_regularizer(...)
: Returns a function that can be used to apply L1 L2 regularizations.
l1_regularizer(...)
: Returns a function that can be used to apply L1 regularization to weights.
l2_regularizer(...)
: Returns a function that can be used to apply L2 regularization to weights.
layer_norm(...)
: Adds a Layer Normalization layer.
legacy_fully_connected(...)
: Adds the parameters for a fully connected layer and returns the output.
legacy_linear(...)
: partial(func, *args, **keywords) - new function with partial application
legacy_relu(...)
: partial(func, *args, **keywords) - new function with partial application
linear(...)
: partial(func, *args, **keywords) - new function with partial application
make_place_holder_tensors_for_base_features(...)
: Returns placeholder tensors for inference.
max_pool2d(...)
: Adds a 2D Max Pooling op.
max_pool3d(...)
: Adds a 3D Max Pooling op.
maxout(...)
: Adds a maxout op from https://arxiv.org/abs/1302.4389
multi_class_target(...)
: Creates a _TargetColumn for multi class single label classification. (deprecated)
one_hot_column(...)
: Creates an _OneHotColumn
for a one-hot or multi-hot repr in a DNN.
one_hot_encoding(...)
: Transform numeric labels into onehot_labels using tf.one_hot
.
optimize_loss(...)
: Given loss and parameters for optimizer, returns a training op.
parse_feature_columns_from_examples(...)
: Parses tf.Examples to extract tensors for given feature_columns.
parse_feature_columns_from_sequence_examples(...)
: Parses tf.SequenceExamples to extract tensors for given FeatureColumn
s.
real_valued_column(...)
: Creates a _RealValuedColumn
for dense numeric data.
recompute_grad(...)
: Decorator that recomputes the function on the backwards pass.
regression_target(...)
: Creates a _TargetColumn for linear regression. (deprecated)
relu(...)
: partial(func, *args, **keywords) - new function with partial application
relu6(...)
: partial(func, *args, **keywords) - new function with partial application
repeat(...)
: Applies the same layer with the same arguments repeatedly.
rev_block(...)
: A block of reversible residual layers.
safe_embedding_lookup_sparse(...)
: Lookup embedding results, accounting for invalid IDs and empty features.
scale_gradient(...)
: _OverloadedFunction encapsulates an overloaded function.
scattered_embedding_column(...)
: Creates an embedding column of a sparse feature using parameter hashing.
separable_conv2d(...)
: Adds a depth-separable 2D convolution with optional batch_norm layer.
separable_convolution2d(...)
: Adds a depth-separable 2D convolution with optional batch_norm layer.
sequence_input_from_feature_columns(...)
: Builds inputs for sequence models from FeatureColumn
s. (experimental)
sequence_to_images(...)
: Convert a batch of sequences into a batch of images.
shared_embedding_columns(...)
: Creates a list of _EmbeddingColumn
sharing the same embedding.
softmax(...)
: Performs softmax on Nth dimension of N-dimensional logit tensor.
sparse_column_with_hash_bucket(...)
: Creates a _SparseColumn with hashed bucket configuration.
sparse_column_with_integerized_feature(...)
: Creates an integerized _SparseColumn.
sparse_column_with_keys(...)
: Creates a _SparseColumn with keys.
sparse_column_with_vocabulary_file(...)
: Creates a _SparseColumn with vocabulary file configuration.
spatial_softmax(...)
: Computes the spatial softmax of a convolutional feature map.
stack(...)
: Builds a stack of layers by applying layer repeatedly using stack_args.
sum_regularizer(...)
: Returns a function that applies the sum of multiple regularizers.
summarize_activation(...)
: Summarize an activation.
summarize_activations(...)
: Summarize activations, using summarize_activation
to summarize.
summarize_collection(...)
: Summarize a graph collection of tensors, possibly filtered by name.
summarize_tensor(...)
: Summarize a tensor using a suitable summary type.
summarize_tensors(...)
: Summarize a set of tensors.
transform_features(...)
: Returns transformed features based on features columns passed in.
unit_norm(...)
: Normalizes the given input across the specified dimension to unit length.
variance_scaling_initializer(...)
: Returns an initializer that generates tensors without scaling variance.
weighted_sparse_column(...)
: Creates a _SparseColumn by combining sparse_id_column with a weight column.
weighted_sum_from_feature_columns(...)
: A tf.contrib.layers style linear prediction builder based on FeatureColumn.
xavier_initializer(...)
: Returns an initializer performing "Xavier" initialization for weights.
xavier_initializer_conv2d(...)
: Returns an initializer performing "Xavier" initialization for weights.