Skip to content
Snippets Groups Projects
Unverified Commit b1fd6890 authored by zhuqi's avatar zhuqi Committed by GitHub
Browse files

Merge pull request #35 from ConvLab/nlu 

Add delex function; NLU benckmark result on multiwoz21, tm1, tm2, tm3
parents 9f5d85a3 4f7f7ed8
Branches
No related tags found
No related merge requests found
Show whitespace changes
Inline Side-by-side
Showing
with 834 additions and 3 deletions
convlab2/nlu/evaluate_unified_datasets.py
+ 5
− 0
View file @ b1fd6890
......@@ -6,8 +6,10 @@ def evaluate(predict_result):
predict_result = json.load(open(predict_result))
metrics = {x: {'TP':0, 'FP':0, 'FN':0} for x in ['overall', 'binary', 'categorical', 'non-categorical']}
acc = []
for sample in predict_result:
flag = True
for da_type in ['binary', 'categorical', 'non-categorical']:
if da_type == 'binary':
predicts = [(x['intent'], x['domain'], x['slot']) for x in sample['predictions']['dialogue_acts'][da_type]]
......@@ -26,6 +28,8 @@ def evaluate(predict_result):
if ele not in predicts:
metrics['overall']['FN'] += 1
metrics[da_type]['FN'] += 1
flag &= (sorted(predicts)==sorted(labels))
acc.append(flag)
for metric in metrics:
TP = metrics[metric].pop('TP')
......@@ -37,6 +41,7 @@ def evaluate(predict_result):
metrics[metric]['precision'] = precision
metrics[metric]['recall'] = recall
metrics[metric]['f1'] = f1
metrics['accuracy'] = sum(acc)/len(acc)
return metrics
......
convlab2/nlu/jointBERT/README.md
+ 44
− 0
View file @ b1fd6890
......@@ -31,6 +31,50 @@ $ python test.py --config_path path_to_a_config_file
The result (`output.json`) will be saved under the `output_dir` of the config file. Also, it will be zipped as `zipped_model_path` in the config file.
## Performance on unified format datasets
To illustrate that it is easy to use the model for any dataset that in our unified format, we report the performance on several datasets in our unified format. We follow `README.md` and config files in `unified_datasets/` to generate `predictions.json`, then evaluate it using `../evaluate_unified_datasets.py`. Note that we use almost the same hyper-parameters for different datasets, which may not be optimal.
<table>
<thead>
<tr>
<th></th>
<th colspan=2>MultiWOZ 2.1</th>
<th colspan=2>Taskmaster-1</th>
<th colspan=2>Taskmaster-2</th>
<th colspan=2>Taskmaster-3</th>
</tr>
</thead>
<thead>
<tr>
<th>Model</th>
<th>Acc</th><th>F1</th>
<th>Acc</th><th>F1</th>
<th>Acc</th><th>F1</th>
<th>Acc</th><th>F1</th>
</tr>
</thead>
<tbody>
<tr>
<td>BERTNLU</td>
<td>74.5</td><td>85.9</td>
<td>72.8</td><td>50.6</td>
<td>79.2</td><td>70.6</td>
<td>86.1</td><td>81.9</td>
</tr>
<tr>
<td>BERTNLU (context=3)</td>
<td>80.6</td><td>90.3</td>
<td>74.2</td><td>52.7</td>
<td>80.9</td><td>73.3</td>
<td>87.8</td><td>83.8</td>
</tr>
</tbody>
</table>
- Acc: whether all dialogue acts of an utterance are correctly predicted
- F1: F1 measure of the dialogue act predictions over the corpus.
## References
```
......
convlab2/nlu/jointBERT/unified_datasets/configs/multiwoz21_user.json
+ 1
− 1
View file @ b1fd6890
......@@ -2,7 +2,7 @@
"dataset_name": "multiwoz21",
"data_dir": "unified_datasets/data/multiwoz21/user/context_window_size_0",
"output_dir": "unified_datasets/output/multiwoz21/user/context_window_size_0",
"zipped_model_path": "unified_datasets/output/multiwoz21/user/context_window_size_0/bertnlu_unified_multiwoz_user_context0.zip",
"zipped_model_path": "unified_datasets/output/multiwoz21/user/context_window_size_0/bertnlu_unified_multiwoz21_user_context0.zip",
"log_dir": "unified_datasets/output/multiwoz21/user/context_window_size_0/log",
"DEVICE": "cuda:0",
"seed": 2019,
......
convlab2/nlu/jointBERT/unified_datasets/configs/multiwoz21_user_context3.json
+ 1
− 1
View file @ b1fd6890
......@@ -2,7 +2,7 @@
"dataset_name": "multiwoz21",
"data_dir": "unified_datasets/data/multiwoz21/user/context_window_size_3",
"output_dir": "unified_datasets/output/multiwoz21/user/context_window_size_3",
"zipped_model_path": "unified_datasets/output/multiwoz21/user/context_window_size_3/bertnlu_unified_multiwoz_user_context3.zip",
"zipped_model_path": "unified_datasets/output/multiwoz21/user/context_window_size_3/bertnlu_unified_multiwoz21_user_context3.zip",
"log_dir": "unified_datasets/output/multiwoz21/user/context_window_size_3/log",
"DEVICE": "cuda:0",
"seed": 2019,
......
convlab2/nlu/jointBERT/unified_datasets/configs/sgd_user.json 0 → 100755
+ 27
− 0
View file @ b1fd6890
{
"dataset_name": "sgd",
"data_dir": "unified_datasets/data/sgd/user/context_window_size_0",
"output_dir": "unified_datasets/output/sgd/user/context_window_size_0",
"zipped_model_path": "unified_datasets/output/sgd/user/context_window_size_0/bertnlu_unified_sgd_user_context0.zip",
"log_dir": "unified_datasets/output/sgd/user/context_window_size_0/log",
"DEVICE": "cuda:0",
"seed": 2019,
"cut_sen_len": 40,
"use_bert_tokenizer": true,
"context_window_size": 0,
"model": {
"finetune": true,
"context": false,
"context_grad": false,
"pretrained_weights": "bert-base-uncased",
"check_step": 1000,
"max_step": 10000,
"batch_size": 128,
"learning_rate": 1e-4,
"adam_epsilon": 1e-8,
"warmup_steps": 0,
"weight_decay": 0.0,
"dropout": 0.1,
"hidden_units": 768
}
}
\ No newline at end of file
convlab2/nlu/jointBERT/unified_datasets/configs/sgd_user_context3.json 0 → 100755
+ 27
− 0
View file @ b1fd6890
{
"dataset_name": "sgd",
"data_dir": "unified_datasets/data/sgd/user/context_window_size_3",
"output_dir": "unified_datasets/output/sgd/user/context_window_size_3",
"zipped_model_path": "unified_datasets/output/sgd/user/context_window_size_3/bertnlu_unified_sgd_user_context3.zip",
"log_dir": "unified_datasets/output/sgd/user/context_window_size_3/log",
"DEVICE": "cuda:0",
"seed": 2019,
"cut_sen_len": 40,
"use_bert_tokenizer": true,
"context_window_size": 3,
"model": {
"finetune": true,
"context": true,
"context_grad": true,
"pretrained_weights": "bert-base-uncased",
"check_step": 1000,
"max_step": 10000,
"batch_size": 128,
"learning_rate": 1e-4,
"adam_epsilon": 1e-8,
"warmup_steps": 0,
"weight_decay": 0.0,
"dropout": 0.1,
"hidden_units": 1536
}
}
\ No newline at end of file
convlab2/nlu/jointBERT/unified_datasets/configs/tm1_user.json 0 → 100755
+ 27
− 0
View file @ b1fd6890
{
"dataset_name": "tm1",
"data_dir": "unified_datasets/data/tm1/user/context_window_size_0",
"output_dir": "unified_datasets/output/tm1/user/context_window_size_0",
"zipped_model_path": "unified_datasets/output/tm1/user/context_window_size_0/bertnlu_unified_tm1_user_context0.zip",
"log_dir": "unified_datasets/output/tm1/user/context_window_size_0/log",
"DEVICE": "cuda:0",
"seed": 2019,
"cut_sen_len": 40,
"use_bert_tokenizer": true,
"context_window_size": 0,
"model": {
"finetune": true,
"context": false,
"context_grad": false,
"pretrained_weights": "bert-base-uncased",
"check_step": 1000,
"max_step": 10000,
"batch_size": 128,
"learning_rate": 1e-4,
"adam_epsilon": 1e-8,
"warmup_steps": 0,
"weight_decay": 0.0,
"dropout": 0.1,
"hidden_units": 768
}
}
\ No newline at end of file
convlab2/nlu/jointBERT/unified_datasets/configs/tm1_user_context3.json 0 → 100755
+ 27
− 0
View file @ b1fd6890
{
"dataset_name": "tm1",
"data_dir": "unified_datasets/data/tm1/user/context_window_size_3",
"output_dir": "unified_datasets/output/tm1/user/context_window_size_3",
"zipped_model_path": "unified_datasets/output/tm1/user/context_window_size_3/bertnlu_unified_tm1_user_context3.zip",
"log_dir": "unified_datasets/output/tm1/user/context_window_size_3/log",
"DEVICE": "cuda:0",
"seed": 2019,
"cut_sen_len": 40,
"use_bert_tokenizer": true,
"context_window_size": 3,
"model": {
"finetune": true,
"context": true,
"context_grad": true,
"pretrained_weights": "bert-base-uncased",
"check_step": 1000,
"max_step": 10000,
"batch_size": 128,
"learning_rate": 1e-4,
"adam_epsilon": 1e-8,
"warmup_steps": 0,
"weight_decay": 0.0,
"dropout": 0.1,
"hidden_units": 1536
}
}
\ No newline at end of file
convlab2/nlu/jointBERT/unified_datasets/configs/tm2_user.json 0 → 100755
+ 27
− 0
View file @ b1fd6890
{
"dataset_name": "tm2",
"data_dir": "unified_datasets/data/tm2/user/context_window_size_0",
"output_dir": "unified_datasets/output/tm2/user/context_window_size_0",
"zipped_model_path": "unified_datasets/output/tm2/user/context_window_size_0/bertnlu_unified_tm2_user_context0.zip",
"log_dir": "unified_datasets/output/tm2/user/context_window_size_0/log",
"DEVICE": "cuda:0",
"seed": 2019,
"cut_sen_len": 40,
"use_bert_tokenizer": true,
"context_window_size": 0,
"model": {
"finetune": true,
"context": false,
"context_grad": false,
"pretrained_weights": "bert-base-uncased",
"check_step": 1000,
"max_step": 10000,
"batch_size": 128,
"learning_rate": 1e-4,
"adam_epsilon": 1e-8,
"warmup_steps": 0,
"weight_decay": 0.0,
"dropout": 0.1,
"hidden_units": 768
}
}
\ No newline at end of file
convlab2/nlu/jointBERT/unified_datasets/configs/tm2_user_context3.json 0 → 100755
+ 27
− 0
View file @ b1fd6890
{
"dataset_name": "tm2",
"data_dir": "unified_datasets/data/tm2/user/context_window_size_3",
"output_dir": "unified_datasets/output/tm2/user/context_window_size_3",
"zipped_model_path": "unified_datasets/output/tm2/user/context_window_size_3/bertnlu_unified_tm2_user_context3.zip",
"log_dir": "unified_datasets/output/tm2/user/context_window_size_3/log",
"DEVICE": "cuda:0",
"seed": 2019,
"cut_sen_len": 40,
"use_bert_tokenizer": true,
"context_window_size": 3,
"model": {
"finetune": true,
"context": true,
"context_grad": true,
"pretrained_weights": "bert-base-uncased",
"check_step": 1000,
"max_step": 10000,
"batch_size": 128,
"learning_rate": 1e-4,
"adam_epsilon": 1e-8,
"warmup_steps": 0,
"weight_decay": 0.0,
"dropout": 0.1,
"hidden_units": 1536
}
}
\ No newline at end of file
convlab2/nlu/jointBERT/unified_datasets/configs/tm3_user.json 0 → 100755
+ 27
− 0
View file @ b1fd6890
{
"dataset_name": "tm3",
"data_dir": "unified_datasets/data/tm3/user/context_window_size_0",
"output_dir": "unified_datasets/output/tm3/user/context_window_size_0",
"zipped_model_path": "unified_datasets/output/tm3/user/context_window_size_0/bertnlu_unified_tm3_user_context0.zip",
"log_dir": "unified_datasets/output/tm3/user/context_window_size_0/log",
"DEVICE": "cuda:0",
"seed": 2019,
"cut_sen_len": 40,
"use_bert_tokenizer": true,
"context_window_size": 0,
"model": {
"finetune": true,
"context": false,
"context_grad": false,
"pretrained_weights": "bert-base-uncased",
"check_step": 1000,
"max_step": 10000,
"batch_size": 128,
"learning_rate": 1e-4,
"adam_epsilon": 1e-8,
"warmup_steps": 0,
"weight_decay": 0.0,
"dropout": 0.1,
"hidden_units": 768
}
}
\ No newline at end of file
convlab2/nlu/jointBERT/unified_datasets/configs/tm3_user_context3.json 0 → 100755
+ 27
− 0
View file @ b1fd6890
{
"dataset_name": "tm3",
"data_dir": "unified_datasets/data/tm3/user/context_window_size_3",
"output_dir": "unified_datasets/output/tm3/user/context_window_size_3",
"zipped_model_path": "unified_datasets/output/tm3/user/context_window_size_3/bertnlu_unified_tm3_user_context3.zip",
"log_dir": "unified_datasets/output/tm3/user/context_window_size_3/log",
"DEVICE": "cuda:0",
"seed": 2019,
"cut_sen_len": 40,
"use_bert_tokenizer": true,
"context_window_size": 3,
"model": {
"finetune": true,
"context": true,
"context_grad": true,
"pretrained_weights": "bert-base-uncased",
"check_step": 1000,
"max_step": 20000,
"batch_size": 64,
"learning_rate": 1e-4,
"adam_epsilon": 1e-8,
"warmup_steps": 0,
"weight_decay": 0.0,
"dropout": 0.1,
"hidden_units": 1536
}
}
\ No newline at end of file
convlab2/nlu/milu/README.md
+ 44
− 0
View file @ b1fd6890
......@@ -43,6 +43,50 @@ Note that the config file is different from the above. You should set:
## Predict
See `nlu.py` under `multiwoz` and `unified_datasets` directories.
## Performance on unified format datasets
To illustrate that it is easy to use the model for any dataset that in our unified format, we report the performance on several datasets in our unified format. We follow `README.md` and config files in `unified_datasets/` to generate `predictions.json`, then evaluate it using `../evaluate_unified_datasets.py`. Note that we use almost the same hyper-parameters for different datasets, which may not be optimal.
<table>
<thead>
<tr>
<th></th>
<th colspan=2>MultiWOZ 2.1</th>
<th colspan=2>Taskmaster-1</th>
<th colspan=2>Taskmaster-2</th>
<th colspan=2>Taskmaster-3</th>
</tr>
</thead>
<thead>
<tr>
<th>Model</th>
<th>Acc</th><th>F1</th>
<th>Acc</th><th>F1</th>
<th>Acc</th><th>F1</th>
<th>Acc</th><th>F1</th>
</tr>
</thead>
<tbody>
<tr>
<td>MILU</td>
<td>72.9</td><td>85.2</td>
<td>72.9</td><td>49.2</td>
<td>79.1</td><td>68.7</td>
<td>85.4</td><td>80.3</td>
</tr>
<tr>
<td>MILU (context=3)</td>
<td>76.6</td><td>87.9</td>
<td>72.4</td><td>48.5</td>
<td>78.9</td><td>68.4</td>
<td>85.1</td><td>80.1</td>
</tr>
</tbody>
</table>
- Acc: whether all dialogue acts of an utterance are correctly predicted
- F1: F1 measure of the dialogue act predictions over the corpus.
## References
```
@inproceedings{lee2019convlab,
......
convlab2/nlu/milu/dataset_reader.py
+ 2
− 0
View file @ b1fd6890
......@@ -76,6 +76,8 @@ class MILUDatasetReader(DatasetReader):
for sample in data:
utterance = sample['utterance']
if len(utterance) == 0:
continue
sentences = self._sent_tokenizer.tokenize(utterance)
sent_spans = self._sent_tokenizer.span_tokenize(utterance)
tokens = [token for sent in sentences for token in self._word_tokenizer.tokenize(sent)]
......
convlab2/nlu/milu/unified_datasets/configs/multiwoz21_user.jsonnet
+ 1
− 1
View file @ b1fd6890
......@@ -98,7 +98,7 @@
"num_epochs": 40,
"grad_norm": 5.0,
"patience": 75,
"cuda_device": 4
"cuda_device": 0
},
"evaluate_on_test": true
}
convlab2/nlu/milu/unified_datasets/configs/sgd_user.jsonnet 0 → 100755
+ 104
− 0
View file @ b1fd6890
{
"dataset_reader": {
"type": "milu",
"token_indexers": {
"tokens": {
"type": "single_id",
"lowercase_tokens": true
},
"token_characters": {
"type": "characters",
"min_padding_length": 3
},
},
"context_size": 0,
"agent": "user",
"use_unified_datasets": true,
"dataset_name": "sgd",
"random_context_size": false
},
"train_data_path": "train",
"validation_data_path": "validation",
"test_data_path": "test",
"model": {
"type": "milu",
"label_encoding": "BIO",
"use_unified_datasets": true,
"dropout": 0.3,
"include_start_end_transitions": false,
"text_field_embedder": {
"token_embedders": {
"tokens": {
"type": "embedding",
"embedding_dim": 50,
"pretrained_file": "https://s3-us-west-2.amazonaws.com/allennlp/datasets/glove/glove.6B.50d.txt.gz",
"trainable": true
},
"token_characters": {
"type": "character_encoding",
"embedding": {
"embedding_dim": 16
},
"encoder": {
"type": "cnn",
"embedding_dim": 16,
"num_filters": 128,
"ngram_filter_sizes": [3],
"conv_layer_activation": "relu"
}
}
}
},
"encoder": {
"type": "lstm",
"input_size": 178,
"hidden_size": 200,
"num_layers": 1,
"dropout": 0.5,
"bidirectional": true
},
"intent_encoder": {
"type": "lstm",
"input_size": 400,
"hidden_size": 200,
"num_layers": 1,
"dropout": 0.5,
"bidirectional": true
},
"attention": {
"type": "bilinear",
"vector_dim": 400,
"matrix_dim": 400
},
"context_for_intent": true,
"context_for_tag": false,
"attention_for_intent": false,
"attention_for_tag": false,
"regularizer": [
[
"scalar_parameters",
{
"type": "l2",
"alpha": 0.1
}
]
]
},
"iterator": {
"type": "basic",
"batch_size": 64
},
"trainer": {
"optimizer": {
"type": "adam",
"lr": 0.001
},
"validation_metric": "+f1-measure",
"num_serialized_models_to_keep": 3,
"num_epochs": 40,
"grad_norm": 5.0,
"patience": 75,
"cuda_device": 0
},
"evaluate_on_test": true
}
convlab2/nlu/milu/unified_datasets/configs/sgd_user_context3.jsonnet 0 → 100755
+ 104
− 0
View file @ b1fd6890
{
"dataset_reader": {
"type": "milu",
"token_indexers": {
"tokens": {
"type": "single_id",
"lowercase_tokens": true
},
"token_characters": {
"type": "characters",
"min_padding_length": 3
},
},
"context_size": 3,
"agent": "user",
"use_unified_datasets": true,
"dataset_name": "sgd",
"random_context_size": false
},
"train_data_path": "train",
"validation_data_path": "validation",
"test_data_path": "test",
"model": {
"type": "milu",
"label_encoding": "BIO",
"use_unified_datasets": true,
"dropout": 0.3,
"include_start_end_transitions": false,
"text_field_embedder": {
"token_embedders": {
"tokens": {
"type": "embedding",
"embedding_dim": 50,
"pretrained_file": "https://s3-us-west-2.amazonaws.com/allennlp/datasets/glove/glove.6B.50d.txt.gz",
"trainable": true
},
"token_characters": {
"type": "character_encoding",
"embedding": {
"embedding_dim": 16
},
"encoder": {
"type": "cnn",
"embedding_dim": 16,
"num_filters": 128,
"ngram_filter_sizes": [3],
"conv_layer_activation": "relu"
}
}
}
},
"encoder": {
"type": "lstm",
"input_size": 178,
"hidden_size": 200,
"num_layers": 1,
"dropout": 0.5,
"bidirectional": true
},
"intent_encoder": {
"type": "lstm",
"input_size": 400,
"hidden_size": 200,
"num_layers": 1,
"dropout": 0.5,
"bidirectional": true
},
"attention": {
"type": "bilinear",
"vector_dim": 400,
"matrix_dim": 400
},
"context_for_intent": true,
"context_for_tag": false,
"attention_for_intent": false,
"attention_for_tag": false,
"regularizer": [
[
"scalar_parameters",
{
"type": "l2",
"alpha": 0.1
}
]
]
},
"iterator": {
"type": "basic",
"batch_size": 64
},
"trainer": {
"optimizer": {
"type": "adam",
"lr": 0.001
},
"validation_metric": "+f1-measure",
"num_serialized_models_to_keep": 3,
"num_epochs": 40,
"grad_norm": 5.0,
"patience": 75,
"cuda_device": 0
},
"evaluate_on_test": true
}
convlab2/nlu/milu/unified_datasets/configs/tm1_user.jsonnet 0 → 100755
+ 104
− 0
View file @ b1fd6890
{
"dataset_reader": {
"type": "milu",
"token_indexers": {
"tokens": {
"type": "single_id",
"lowercase_tokens": true
},
"token_characters": {
"type": "characters",
"min_padding_length": 3
},
},
"context_size": 0,
"agent": "user",
"use_unified_datasets": true,
"dataset_name": "tm1",
"random_context_size": false
},
"train_data_path": "train",
"validation_data_path": "validation",
"test_data_path": "test",
"model": {
"type": "milu",
"label_encoding": "BIO",
"use_unified_datasets": true,
"dropout": 0.3,
"include_start_end_transitions": false,
"text_field_embedder": {
"token_embedders": {
"tokens": {
"type": "embedding",
"embedding_dim": 50,
"pretrained_file": "https://s3-us-west-2.amazonaws.com/allennlp/datasets/glove/glove.6B.50d.txt.gz",
"trainable": true
},
"token_characters": {
"type": "character_encoding",
"embedding": {
"embedding_dim": 16
},
"encoder": {
"type": "cnn",
"embedding_dim": 16,
"num_filters": 128,
"ngram_filter_sizes": [3],
"conv_layer_activation": "relu"
}
}
}
},
"encoder": {
"type": "lstm",
"input_size": 178,
"hidden_size": 200,
"num_layers": 1,
"dropout": 0.5,
"bidirectional": true
},
"intent_encoder": {
"type": "lstm",
"input_size": 400,
"hidden_size": 200,
"num_layers": 1,
"dropout": 0.5,
"bidirectional": true
},
"attention": {
"type": "bilinear",
"vector_dim": 400,
"matrix_dim": 400
},
"context_for_intent": true,
"context_for_tag": false,
"attention_for_intent": false,
"attention_for_tag": false,
"regularizer": [
[
"scalar_parameters",
{
"type": "l2",
"alpha": 0.1
}
]
]
},
"iterator": {
"type": "basic",
"batch_size": 64
},
"trainer": {
"optimizer": {
"type": "adam",
"lr": 0.001
},
"validation_metric": "+f1-measure",
"num_serialized_models_to_keep": 3,
"num_epochs": 40,
"grad_norm": 5.0,
"patience": 75,
"cuda_device": 0
},
"evaluate_on_test": true
}
convlab2/nlu/milu/unified_datasets/configs/tm1_user_context3.jsonnet 0 → 100755
+ 104
− 0
View file @ b1fd6890
{
"dataset_reader": {
"type": "milu",
"token_indexers": {
"tokens": {
"type": "single_id",
"lowercase_tokens": true
},
"token_characters": {
"type": "characters",
"min_padding_length": 3
},
},
"context_size": 3,
"agent": "user",
"use_unified_datasets": true,
"dataset_name": "tm1",
"random_context_size": false
},
"train_data_path": "train",
"validation_data_path": "validation",
"test_data_path": "test",
"model": {
"type": "milu",
"label_encoding": "BIO",
"use_unified_datasets": true,
"dropout": 0.3,
"include_start_end_transitions": false,
"text_field_embedder": {
"token_embedders": {
"tokens": {
"type": "embedding",
"embedding_dim": 50,
"pretrained_file": "https://s3-us-west-2.amazonaws.com/allennlp/datasets/glove/glove.6B.50d.txt.gz",
"trainable": true
},
"token_characters": {
"type": "character_encoding",
"embedding": {
"embedding_dim": 16
},
"encoder": {
"type": "cnn",
"embedding_dim": 16,
"num_filters": 128,
"ngram_filter_sizes": [3],
"conv_layer_activation": "relu"
}
}
}
},
"encoder": {
"type": "lstm",
"input_size": 178,
"hidden_size": 200,
"num_layers": 1,
"dropout": 0.5,
"bidirectional": true
},
"intent_encoder": {
"type": "lstm",
"input_size": 400,
"hidden_size": 200,
"num_layers": 1,
"dropout": 0.5,
"bidirectional": true
},
"attention": {
"type": "bilinear",
"vector_dim": 400,
"matrix_dim": 400
},
"context_for_intent": true,
"context_for_tag": false,
"attention_for_intent": false,
"attention_for_tag": false,
"regularizer": [
[
"scalar_parameters",
{
"type": "l2",
"alpha": 0.1
}
]
]
},
"iterator": {
"type": "basic",
"batch_size": 64
},
"trainer": {
"optimizer": {
"type": "adam",
"lr": 0.001
},
"validation_metric": "+f1-measure",
"num_serialized_models_to_keep": 3,
"num_epochs": 40,
"grad_norm": 5.0,
"patience": 75,
"cuda_device": 0
},
"evaluate_on_test": true
}
convlab2/nlu/milu/unified_datasets/configs/tm2_user.jsonnet 0 → 100755
+ 104
− 0
View file @ b1fd6890
{
"dataset_reader": {
"type": "milu",
"token_indexers": {
"tokens": {
"type": "single_id",
"lowercase_tokens": true
},
"token_characters": {
"type": "characters",
"min_padding_length": 3
},
},
"context_size": 0,
"agent": "user",
"use_unified_datasets": true,
"dataset_name": "tm2",
"random_context_size": false
},
"train_data_path": "train",
"validation_data_path": "validation",
"test_data_path": "test",
"model": {
"type": "milu",
"label_encoding": "BIO",
"use_unified_datasets": true,
"dropout": 0.3,
"include_start_end_transitions": false,
"text_field_embedder": {
"token_embedders": {
"tokens": {
"type": "embedding",
"embedding_dim": 50,
"pretrained_file": "https://s3-us-west-2.amazonaws.com/allennlp/datasets/glove/glove.6B.50d.txt.gz",
"trainable": true
},
"token_characters": {
"type": "character_encoding",
"embedding": {
"embedding_dim": 16
},
"encoder": {
"type": "cnn",
"embedding_dim": 16,
"num_filters": 128,
"ngram_filter_sizes": [3],
"conv_layer_activation": "relu"
}
}
}
},
"encoder": {
"type": "lstm",
"input_size": 178,
"hidden_size": 200,
"num_layers": 1,
"dropout": 0.5,
"bidirectional": true
},
"intent_encoder": {
"type": "lstm",
"input_size": 400,
"hidden_size": 200,
"num_layers": 1,
"dropout": 0.5,
"bidirectional": true
},
"attention": {
"type": "bilinear",
"vector_dim": 400,
"matrix_dim": 400
},
"context_for_intent": true,
"context_for_tag": false,
"attention_for_intent": false,
"attention_for_tag": false,
"regularizer": [
[
"scalar_parameters",
{
"type": "l2",
"alpha": 0.1
}
]
]
},
"iterator": {
"type": "basic",
"batch_size": 64
},
"trainer": {
"optimizer": {
"type": "adam",
"lr": 0.001
},
"validation_metric": "+f1-measure",
"num_serialized_models_to_keep": 3,
"num_epochs": 40,
"grad_norm": 5.0,
"patience": 75,
"cuda_device": 0
},
"evaluate_on_test": true
}
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment