- 
                Notifications
    
You must be signed in to change notification settings  - Fork 133
 
Mapping Nematus model parameters to S2S
        Roman Grundkiewicz edited this page Sep 1, 2017 
        ·
        1 revision
      
    | S2S Parameter | Value | Nematus Parameter | Explanation | 
|---|---|---|---|
| dim-vocabs | <list> | n_words_src, n_words | Maximum items in vocabulary ordered by rank | 
| dim-emb | int | dim_word | Size of embedding vector | 
| dim-rnn | int | dim | Size of rnn hidden state | 
| enc-type | str | NA | Type of encoder RNN : bidirectional, bi-unidirectional, alternating (s2s) | 
| enc-cell | str | encoder | Type of RNN cell: gru, lstm, tanh (s2s) | 
| enc-cell-depth | int | enc_recurrence_transition_depth | Number of tansitional cells in encoder layers (s2s) | 
| enc-depth | int | enc_depth | Number of encoder layers (s2s) | 
| dec-cell | str | decoder_deep | Type of RNN cell: gru, lstm, tanh (s2s) | 
| dec-cell-base-depth | int | dec_base_recurrence_transition_depth | Number of tansitional cells in first decoder layer (s2s) | 
| dec-cell-high-depth | int | dec_high_recurrence_transition_depth | Number of tansitional cells in next decoder layers (s2s) | 
| dec-depth | int | dec_depth | Number of decoder layers (s2s) | 
| skip | NA | Use skip connections (s2s) | |
| layer-normalization | bool | layer_normalization | Enable layer normalization | 
| best-deep | NA | Use WMT-2017-style deep configuration (s2s) | |
| special-vocab | NA | Model-specific special vocabulary ids | |
| tied-embeddings | bool | tie_decoder_embeddings | Tie target embeddings and output embeddings in output layer | 
| tied-embeddings-src | bool | Tie source and target embeddings | |
| tied-embeddings-all | bool | Tie all embedding layers and output layer |