0.10.3

Core (0.7.1 ==> 0.8.4)

Features

• Add tf.toPixels. (#941)
• Add tf.erf op (#951). Thanks @Lewuathe.
• Add tf.movingAverage. (#963)
• Add tf.resizeNearestNeighbor (#955). Thanks @Lewuathe.
• Add slice ergonomics. (#964). Thanks @ry.
• Finish gradient for tf.pow (#954). Thanks @jgartman.
• Add gradients for tf.batchNormalization4d (#959)
• Finish gradient for tf.matMul (#957). Thanks @jgartman.
• Add tf.logSigmoid and tf.softplus and gradients (#916). Thanks @jgartman.

Breaking changes

• Require shape in tensor2d/3d/4d(flatValues, shape) (#969).
• Align API with TF (#956).

Aligns the backend API and functionality (NaN propagation, dtype strictness, kernel signatures) with TensorFlow Python.

• Remove backend.minPool since TF doesn't have it.
• Remove normRegion param in localResponseNormalization kernel since TF doesn't support it.
• Remove leakyRelu, prelu and preluDer from the backend and implement using higher-level ops, aligning with TF Python.
• Make backend.multinomial take logits instead of probabilities and normalized: boolean param for backwards compatibility.
• argMin and argMax take single axis: number instead of axes: number[]
• Change eluDer(x: T): T signature to eluDer(dy: T, y: T): T to align with TF.
• Change NaN behavior of max/avgPool and conv2d to align with TF.
• Change avgPool out of bounds (padding) behavior to align with TF
• Require indices in oneHot and gather to be of dtype int32

Fixes #195

• Align webgl and cpu backend to be closer to TF (#947).

• Remove NaN support for tensors of dtype bool and int32
• Remove related tests
• Make fromPixels test run only in the browser
• Remove NaN propagation for min/max/argMin/argMax/compare/logical ops to align with TF
• Speed up randomNormal tests
• Minimize usage of isNaN checks in WebGL shaders

Bug fixes

• Implement floordiv for integer division & update CONTRIBUTING.md. (#960)

Performance

• Make CPU matmul 100x faster (#961). Thanks @piscisaureus.

TypeScript

• Modify tf.split signature to preserve rank. (#973). Thanks @iansimon.

Development

• Add yarn "link-local" and yarn "publish-local" (#983).

These commands are better alternatives to yarn link and resolve caching and de-duplication problems:

• yarn publish-local replaces yarn link
• yarn link-local PCK_NAME replaces yarn link PCK_NAME

• Add quotemark rule to tslint. (#975). Thanks @ry.
• Make core tests reusable by other packages (#926).

Documentation

• Add code examples to jsdocs. (#899). Thanks @nbardy.
• Fix typo in tsdoc (#952). Thanks @Lewuathe.
• Fix snippet comments for sum and mean (#946). Thanks @mlajtos.
• Array Ops: Fixes tf.clone documentation (#942). Thanks @manrajgrover.

Misc

• Assert arguments to ops are Tensors. (#967)
• test_util shouldn't depend on jasmine. (#974). Thanks @ry.
• Clean up demo scripts. (#949). Thanks @manrajgrover.
• Fix error message in clipByValue (#953). Thanks @mlajtos.
• Remove deeplearnjs references (#948). Thanks @manrajgrover.
• remove old demos, move benchmarks to integration_tests (#943).
• Rename dl -> tf in unit tests (#944). Thanks @manrajgrover.

Layers (0.4.0 ==> 0.5.3)

Features

• Support training of BatchNormalization layer (#139)
• Expose tf.metrics; Add binaryCrossentropy metric (#144)
• Export Reshape layer as tf.layers.reshape (#134).

Breaking changes

• Typo: Corrected typo thresohldedReLU -> thresholdedReLU (#136). Thanks @xam-ps.

Bug fixes

• Fix serialization of Sequential (#127)
• Cast indices to int as needed. (#128).
• Let Conv1D take [number] for dilationRate (#125)

Performance

TypeScript

• Export Shape type alias: tf.Shape (#142). Thanks @ry.

Development

• Improve DEVELOPMENT.md (#148).

• Add yarn link-local and yarn publish-local to substitute yarn link and avoid problems with duplication of packages and caching

• Add quotemark to tslint. (#143). Thanks @ry.
• Update switch-tfjs-core-version.sh (switch-deeplearn-version.sh) (#138)

Documentation

• add @doc to getLayer (#145)
• Add some print-outs to a code snippet (#133).

Misc

• Remove underscore as a dependency (#129)
• Issues Template: Remove label issues from template (#124). Thanks @manrajgrover.

0.9.1

Bug fix

• Layers: Fix bug in which Conv1D Layer cannot take an Array of a single number as dilationRate.

0.9.0

New features

• Core ops: tf.rsqrt, tf.mod, tf.losses.absoluteDifference, tf.losses.computeWeightedLoss, tf.losses.squaredDifference, tf.sign, tf.floor, tf.reciprocal, tf.round, tf.asinh, tf.acosh, tf.atanh, tf.atan2.
• Core: Centered RMSProp now supported.
• Core: logicalNot, cast supports chaining.
• Core: tf.stack and tf.concat allow a single tensor.
• Core: Shader programs now work in GLSL 4.1.
• Core: loadWeights now supports RequestInit.
• Layers: Support dilation in Conv1D and Conv2D layers.
• Layers: Add SeparableConv2D Layer (tf.layers.separableConv2d).

Bug fixes

• Layers: Map weights to uniquefied tensor names during model loading, to prevent error on repeated loading of the same model with tf.loadModel() in the same browser page load.
• Layers: Change class registry to AveragePooling(1|2)D and add symbol aliases to exports.

Acknowledgments to external contributors:

@manrajgrover for tf.exmp1, tf.squaredDifference, tf.rsqrt, tf.mod, absolute difference loss, tf.sign, fixing gradient for tf.floor, tf.reciprocal, logicalNot chaining support, fixing documentation, adding info to issue template.
@Lewuathe for centered RMSProp, unary ops benchmarks, tf.round. @jgartman for inverse trig hyperbolic functions and tf.atan2. @adarob for tf.split, 1-tensor tf.stack and tf.concat. @adarabi for fixing GLSL shader compiler for glsl 4.1.

0.8.0

New features

• Layers: Add ZeroPadding2D layer.

Bug fixes

• Layers: Fix duplicate weight names in Conv2DTranspose layer.

Acknowledgments to external contributors

@adarabi for shader syntax, @Lewuathe for work on tests, @manrajgrover for new ops, @jgartman for new ops

0.7.0

New features

• Layers: Added some advanced activation layer types: LeakyReLU, ELU, ThresholdedReLU.
• Layers: Added Conv2DTranspose layer. Inference is supported; training (backpropagation) is not supported yet.

Bug fixes

• Layers: Fix a bug related to optimizer constructor names in minimized bundle.

Acknowledgments to external contributors:

bringtree@