-
Notifications
You must be signed in to change notification settings - Fork 561
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
feat: LSTM over torch , preliminary internal graph representation
preliminary steps clean dot output allocation of torch modules + forward simple forward ut ok lstm from proto trained by torch ok csvts torch connector, 1 working learning of csvts with torch backend gpu ok make lstm statefull if needed fix iterations display number cleanup realloc computations fix name collision when compiled w/ TORCH and w/caffe can pass labels either at creation of at train call add some headers for some versions remove -liomp if cuda version fix ambiguous var name load correct file do not double init params load directly models on correct device torch timeseries prediction ok fix test wrt correct init do not regenate protoxt if already present : allows not to give net definition at predict time reload params after realloc load only if weights are present small changes as for beniz review changes for sileht and louisj reviews
- Loading branch information
Showing
32 changed files
with
2,977 additions
and
236 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,227 @@ | ||
| name: "recurrent" | ||
| layer { | ||
| name: "inputl" | ||
| type: "MemoryData" | ||
| top: "data" | ||
| top: "label" | ||
| include { | ||
| phase: TRAIN | ||
| } | ||
| memory_data_param { | ||
| batch_size: 1 | ||
| channels: 50 | ||
| height: 9 | ||
| width: 1 | ||
| } | ||
| } | ||
| layer { | ||
| name: "inputl" | ||
| type: "MemoryData" | ||
| top: "data" | ||
| top: "label" | ||
| include { | ||
| phase: TEST | ||
| } | ||
| memory_data_param { | ||
| batch_size: 1 | ||
| channels: 50 | ||
| height: 9 | ||
| width: 1 | ||
| } | ||
| } | ||
| layer { | ||
| name: "permute_T_N_data" | ||
| type: "Permute" | ||
| bottom: "data" | ||
| top: "permuted_data" | ||
| permute_param { | ||
| order: 1 | ||
| order: 0 | ||
| order: 2 | ||
| order: 3 | ||
| } | ||
| } | ||
| layer { | ||
| name: "slice_timeseries" | ||
| type: "Slice" | ||
| bottom: "permuted_data" | ||
| top: "cont_seq_unshaped" | ||
| top: "target_seq" | ||
| top: "input_seq" | ||
| slice_param { | ||
| slice_point: 1 | ||
| slice_point: 4 | ||
| axis: 2 | ||
| } | ||
| } | ||
| layer { | ||
| name: "shape_cont_seq" | ||
| type: "Flatten" | ||
| bottom: "cont_seq_unshaped" | ||
| top: "cont_seq" | ||
| flatten_param { | ||
| axis: 1 | ||
| } | ||
| } | ||
| layer { | ||
| name: "LSTM0" | ||
| type: "LSTM" | ||
| bottom: "input_seq" | ||
| bottom: "cont_seq" | ||
| top: "LSTM_0" | ||
| recurrent_param { | ||
| num_output: 50 | ||
| weight_filler { | ||
| type: "uniform" | ||
| min: -0.14142136 | ||
| max: 0.14142136 | ||
| } | ||
| bias_filler { | ||
| type: "uniform" | ||
| min: -0.14142136 | ||
| max: 0.14142136 | ||
| } | ||
| } | ||
| } | ||
| layer { | ||
| name: "LSTM1" | ||
| type: "LSTM" | ||
| bottom: "LSTM_0" | ||
| bottom: "cont_seq" | ||
| top: "LSTM_1" | ||
| recurrent_param { | ||
| num_output: 50 | ||
| weight_filler { | ||
| type: "uniform" | ||
| min: -0.14142136 | ||
| max: 0.14142136 | ||
| } | ||
| bias_filler { | ||
| type: "uniform" | ||
| min: -0.14142136 | ||
| max: 0.14142136 | ||
| } | ||
| } | ||
| } | ||
| layer { | ||
| name: "affine_2" | ||
| type: "InnerProduct" | ||
| bottom: "LSTM_1" | ||
| top: "rnn_pred" | ||
| inner_product_param { | ||
| num_output: 3 | ||
| weight_filler { | ||
| type: "uniform" | ||
| min: -0.14142136 | ||
| max: 0.14142136 | ||
| } | ||
| bias_filler { | ||
| type: "uniform" | ||
| min: -0.14142136 | ||
| max: 0.14142136 | ||
| } | ||
| axis: 2 | ||
| } | ||
| } | ||
| layer { | ||
| name: "permute_T_N_rnn_pred" | ||
| type: "Permute" | ||
| bottom: "rnn_pred" | ||
| top: "permuted_rnn_pred" | ||
| include { | ||
| phase: TRAIN | ||
| } | ||
| permute_param { | ||
| order: 1 | ||
| order: 0 | ||
| order: 2 | ||
| } | ||
| } | ||
| layer { | ||
| name: "permute_T_N_target_seq" | ||
| type: "Permute" | ||
| bottom: "target_seq" | ||
| top: "permuted_target_seq" | ||
| include { | ||
| phase: TRAIN | ||
| } | ||
| permute_param { | ||
| order: 1 | ||
| order: 0 | ||
| order: 2 | ||
| } | ||
| } | ||
| layer { | ||
| name: "Target_Seq_Dim" | ||
| type: "Flatten" | ||
| bottom: "permuted_target_seq" | ||
| top: "permuted_target_seq_flattened" | ||
| include { | ||
| phase: TRAIN | ||
| } | ||
| flatten_param { | ||
| axis: 2 | ||
| } | ||
| } | ||
| layer { | ||
| name: "Loss_Sum_Layer" | ||
| type: "Eltwise" | ||
| bottom: "permuted_rnn_pred" | ||
| bottom: "permuted_target_seq_flattened" | ||
| top: "difference" | ||
| include { | ||
| phase: TRAIN | ||
| } | ||
| eltwise_param { | ||
| operation: SUM | ||
| coeff: 1 | ||
| coeff: -1 | ||
| } | ||
| } | ||
| layer { | ||
| name: "Loss_Reduction" | ||
| type: "Reduction" | ||
| bottom: "difference" | ||
| top: "summed_difference" | ||
| include { | ||
| phase: TRAIN | ||
| } | ||
| reduction_param { | ||
| operation: ASUM | ||
| axis: 1 | ||
| } | ||
| } | ||
| layer { | ||
| name: "Loss_Scale" | ||
| type: "Scale" | ||
| bottom: "summed_difference" | ||
| top: "scaled_difference" | ||
| param { | ||
| lr_mult: 0 | ||
| decay_mult: 0 | ||
| } | ||
| include { | ||
| phase: TRAIN | ||
| } | ||
| scale_param { | ||
| axis: 0 | ||
| filler { | ||
| type: "constant" | ||
| value: 1 | ||
| } | ||
| bias_term: false | ||
| } | ||
| } | ||
| layer { | ||
| name: "Loss_Reduction_Batch" | ||
| type: "Reduction" | ||
| bottom: "scaled_difference" | ||
| top: "loss" | ||
| loss_weight: 1 | ||
| include { | ||
| phase: TRAIN | ||
| } | ||
| reduction_param { | ||
| operation: SUM | ||
| } | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.