Operators missing for MXNet backend

[sandeep-krishnamurthy]

Variables and Placeholders

• Support for constraints in Keras variables and Placeholders

Update Operators

• update
• update_add
• update_sub

Graph Manipulations

• gradients

Layers

• Embedding
• noise.GaussianNoise
• noise.GaussianDropout
• noise.AlphaDropout
• ConvLSTM2D

RNNs

• rnn

CNNs

• conv1d
• conv2d_transpose
• separable_conv2d
• depthwise_conv2d
• conv3d
• conv3d_transpose
• local_conv1d
• local_conv2d
• pooling with SAME mode
• conv1d with CAUSAL mode
• separable_conv1D

Higher Order Functions

• map_fn
• foldl
• foldr

Sparse Tensors

• Sparse tensors are supported
• sparse sum
• sparse mean
• sparse concat
• sparse dot
• sparse embedding

NN Operators

• sparse_categorical_crossentropy

Optimizers

• Adam_AMSGrad

Others

• truncated_normal
• cumsum
• cumprod
• logsumexp
• stack
• slice
• ctc
• module - Modulo operator is not supported #37
• gather operator does not work with Embedding Layer - https://github.com/awslabs/keras-apache-mxnet/issues/6300000000000
• Pool2D with SAME mode.
• Depthwise and Separable Conv2D with multiplier != 1 and stride != 1
• Partial Loss is not supported. See here for more details - pytest tests/keras/engine/test_training.py -k "test_model_with_partial_loss"
• External Loss is not supported. See here for more details - pytest tests/keras/engine/test_training.py -k "test_model_with_external_loss"
• Does not support clone model

[Robchio]

Sorry for the obvious question - what's the meaning of a 'checked box' vs. an empty box?

Has 'checked' been implemented? For example, does keras-apache-mxnet support conv1d model layers or not?

[kalyc]

checked means implemented. So all of above operators containing a check mark beside their name are supported