TensorFlow.js 0.14.1 release!

[Nikhil Thorat]

High-level additions

• tfjs-data has launched with: tf.data.array, tf.data.csv, tf.data.zip and chainable transformation methods on a tf.data.Dataset.
• New ops: tf.spectral.fft, tf.spectral.rfft, tf.spectral.irfft, tf.spectral.ifft, tf.conv3d, tf.math.confusionMatrix
• New ops supported by the tfjs-converter: FloorMod, BatchMatMul, CropAndResize, SplitV, Atan2, LeakyRelu, LogSoftmax, SparseToDense, ListDiff, SparseToDense, Prod, FFT, IFFT, RFFT, IRFFT, ScatterND, GatherND
• New layers in tfjs-layers: tf.layers.dot, tf.layers.relu
• Added new method tf.Model.fitDataset which allows passing a tf.data.Dataset from tf.data to a fit call.

Breaking changes

• Implementation of BatchNormalization training code in layers now aligns with tf.keras (#385).

See the full release notes at: https://github.com/tensorflow/tfjs/releases/tag/v0.14.1