Skip to content
This repository has been archived by the owner on Dec 29, 2022. It is now read-only.

Latest commit

 

History

History
75 lines (48 loc) · 2.74 KB

CHANGELIST.md

File metadata and controls

75 lines (48 loc) · 2.74 KB

0.7.3

  1. Maintenance release w/ some deprecation notice fixes. Note: this may change the names of the summaries.

0.7.2

  1. Maintenance release w/ change to project pip dependencies to better support GPU builds.

0.7.1

General

  1. Changed weights to init and bias_init to bias and made these support initialization functions or Tensors.
  2. Added parameter_modifier. These are functions that are applied after creating a Variable, but before it is used in the graph. They allow you to apply a function like normalization or drop connect to the graph. See pt.parameters for details.
  3. Added support for directly chaining many useful TensorFlow functions. See pretty_tensor_methods.py for details. Note: when a function is removed from tf (e.g. complex_abs), it will be removed here.
  4. Changed internal calls to TF to comply with API changes.
  5. Internally changed the name of the first parameter to be more consistent. This should not be user visible since it is the variable to the left of the '.'.

Losses

  1. Added per_output_weights to binary_cross_entropy_with_logits and that allow you to weight the loss from classes and examples.
  2. Added sparse_cross_entropy to efficiently calculate the loss of when you have a vector of 1 hot labels as indices (tf.int32/tf.int64). Also added evaluate_classifier_sparse.
  3. Fixed softmax_classifier_with_sampled_loss to support specified parameters and parameter modification.
  4. Standardized on num_classes and changed the parameter name in softmax_classifier accordingly.

Optimizer

  1. Added clip_gradients_by_norm to apply_optimizer.

Images

  1. Added a differentiable sampling method for images called bilinear_sampling.

0.6.2

Add Depthwise Convolution

Batch Normalization

  1. Make Batch Normalization work with arbitrary dimensionality.
  2. Allow passing through arguments to BN using a namedtuple.
  3. Add BN default values.
  4. Remove requirement to use with_update_ops to make BN accumulate values for inference.

0.6.0

  1. Adding scoped control of summary creation.
  2. Scoped variable collections.
  3. Can initialize variables from literals.
  4. Fixed operators -- Sequential's plus no longer has side effects.
  5. Operators now work on Pretty Tensors that contain lists.

Note: (4) may be breaking!

0.5.3

  1. Fixing tutorials (thanks jkahn!)
  2. Adding a precicion and recall evaluation.
  3. Various bug fixes.

Tested on TF 0.7.1

0.5.2

  1. Various bug fixes
  2. Reordered the arguments to a better positional order.
  3. Added a length argument to recurrent networks to support short circuiting.
  4. Improvements to reshape.
  5. Python 3 support.

0.5.0

Initial Release