Not supported keras/examples with MXNet backend

[roywei]

Checked examples are tested to be working with MXNet backend Not supported examples have clear error message specifying the exact functionality MXNet does not support yet

• [x] addition_rnn.py
• [x] antirectifier.py
• [x] babi_memnn.py
• [x] babi_rnn.py
• [x] cifar10_cnn.py
• [ ] cifar10_cnn_capsule.py [Custom CNN infer shape Not supported]
• [ ] cifar10_cnn_tfaugment2d.py [TF specific Not supported]
• [x] cifar10_resnet.py
• [ ] conv_filter_visualization.py [Gradient Not supported]
• [ ] conv_lstm.py [Conv2DLSTM Not supported]
• [ ] deep_dream.py [Gradient Not supported]
• [ ] image_ocr.py [CTC Not supported]
• [x] imdb_bidirectional_lstm.py
• [x] imdb_cnn.py
• [x] imdb_cnn_lstm.py
• [x] imdb_fasttext.py
• [x] imdb_lstm.py
• [x] lstm_seq2seq.py
• [x] lstm_seq2seq_restore.py
• [x] lstm_stateful.py
• [x] lstm_text_generation.py
• [x] mnist_acgan.py
• [x] mnist_cnn.py
• [ ] mnist_dataset_api.py [TF Specific Not supported]
• [x] mnist_denoising_autoencoder.py
• [x] mnist_hierarchical_rnn.py
• [x] mnist_irnn.py
• [x] mnist_mlp.py
• [x] mnist_net2net.py
• [x] mnist_siamese.py
• [x] mnist_sklearn_wrapper.py
• [ ] mnist_swwae.py [Gradient Not supported]
• [ ] mnist_tfrecord.py [TF Specific Not supported]
• [x] mnist_transfer_cnn.py
• [ ] neural_doodle.py [Gradient Not supported]
• [ ] neural_style_transfer.py [Gradient Not supported]
• [x] pretrained_word_embeddings.py
• [x] reuters_mlp.py
• [ ] reuters_mlp_relu_vs_selu.py
• [ ] variational_autoencoder.py [Custom Loss Not supported]
• [ ] variational_autoencoder_deconv.py [Custom Loss Not supported]

[sandeep-krishnamurthy]

@roywei - I was unable to run seq2seq example.

[roywei]

lstm_seq2seq.py is working in latest master

[kalyc]

As discussed offline, babi_memnn.py & mnist_acgan.py uses sparse_categorical_cross_entropy which is independent of sparse tensor support. Maybe we should label that more clearly above?

[roywei]

@kalyc updated, we should treat is as a normal operator, but still investigate whether it take sparse tensors just like other normal ops.

[roywei]

With #148 , two more examples enabled. (babi_memnn.py and mnist_acgan.py)

[kalyc]

With the PR #153, reuters_mlp_relu_vs_selu.py example has been disabled as AlphaDropout is not supported by MXNet