Policy Manager and NLG modules for MultiWOZ

CHANGELOG.md

@@ -82,6 +82,8 @@ To release a new version, please update the changelog as followed:
 - Added documentation for 8 kHz model ([PR #632](https://github.com/NVIDIA/NeMo/pull/632)) - @jbalam-nv
 - The Neural Graph is a high-level abstract concept empowering the users to build graphs consisting of many, interconnected Neural Modules. A user in his/her application can build any number of graphs, potentially spanning over the same modules. The import/export options combined with the lightweight API make Neural Graphs a perfect tool for rapid prototyping and experimentation. ([PR #413](https://github.com/NVIDIA/NeMo/pull/413)) - @tkornuta-nvidia
 - Created the NeMo CV collection, added  MNIST and CIFAR10 thin datalayers, implemented/ported several general usage trainable and non-trainable modules, added several new ElementTypes ([PR #654](https://github.com/NVIDIA/NeMo/pull/654)) - @tkornuta-nvidia
+- Added SGD dataset and SGD model baseline ([PR #612](https://github.com/NVIDIA/NeMo/pull/612)) - @ekmb
+- Policy Manager and Natural Language Generation Modules for MultiWOZ added ([PR #691](https://github.com/NVIDIA/NeMo/pull/691)) - @ekmb
 
 
 ### Changed

Jenkinsfile

@@ -164,6 +164,13 @@ pipeline {
             sh 'rm -rf examples/nlp/glue_benchmark/glue_output'
           }
         }
+        stage ('TRADE-Rule-based-DPM/NLG') {
+          steps {
+            sh 'cd examples/nlp/dialogue_state_tracking && python rule_based_policy_multiwoz.py --data_dir /home/TestData/nlp/multiwoz2.1/pm_nlg \
+            --encoder_ckpt /home/TestData/nlp/multiwoz2.1/pm_nlg/ckpts/EncoderRNN-EPOCH-10.pt \
+            --decoder_ckpt /home/TestData/nlp/multiwoz2.1/pm_nlg/ckpts/TRADEGenerator-EPOCH-10.pt'
+          }
+        }
       }
     }
 

docs/sources/source/nlp/dialogue_state_tracking.rst

@@ -256,6 +256,33 @@ You may find the checkpoints for the trained models on MultiWOZ 2.0 and MultiWOZ
     predict special values for like **don't care** or **none** for the slots. The `process_multiwoz.py`_ script extracts the additional labels from the dataset and `dialogue_state_tracking_trade.py`_ script reports the **Gating Accuracy** as well.
 
 
+Complete dialogue Pipeline with TRADE for MultiWOZ
+--------------------------------------------------
+
+The pre-trained TRADE model, as was mentioned above, is a Dialogue State Tracker (DST) responsible for extracting correct slot-slot_value pairs from the dialogue history. Once the system has this information, \
+the next module called Dialogue Policy Manager (DPM) comes into play.
+This module determines what actions the system should take, given the dialogue state passed from the DST module. For example, the DPM can request additional information from the user or \
+inform the user about possible ways/options to fill out the user's original intent. 
+With the output of the DPM, the final dialogue module called Natural Language Generation (NLG) generates the system's response to the user's utterance.
+
+NeMo provides Rule-based DPM and Rule-based NLG modules (source: `ConvLab-2: An Open-Source Toolkit for Building, Evaluating, and Diagnosing Dialogue Systems <https://github.com/thu-coai/ConvLab-2>`_) \
+to complete the dialogue pipeline based on the TRADE model and MultiWOZ dataset.
+
+To evaluate TRADE's model output and its role in the complete dialogue pipeline, use ``examples/nlp/dialogue_state_tracking\rule_based_policy_multiwoz.py``.
+Before running this script, make sure to download the pre-trained TRADE model checkpoint following the steps above.
+
+.. code-block:: bash
+
+    cd examples/nlp/dialogue_state_tracking
+    python rule_based_policy_multiwoz.py \
+        --data_dir <path to the data> \
+        --encoder_ckpt <path to checkpoint folder>\EncoderRNN.pt \
+        --decoder_ckpt <path to checkpoint folder>\TRADEGenerator.pt \
+        --mode example \
+
+Use ``--mode interactive`` to chat with the system and ``--hide_output`` - to hide the intermediate output of the dialogue modules
+
+
 References
 ----------
 

examples/nlp/dialogue_state_tracking/data/multiwoz/process_multiwoz.py

@@ -48,14 +48,17 @@
 This script can be used to process and import MultiWOZ 2.0 and 2.1 datasets. 
 You may find more information on how to use this example in NeMo's documentation:
 https://nvidia.github.io/NeMo/nlp/dialogue_state_tracking_trade.html
+
+This file contains code artifacts adapted from the original implementation:
+https://github.com/thu-coai/ConvLab-2
 """
 
 import argparse
 import json
 import os
 import re
-import shutil
-from os.path import exists
+from os.path import exists, expanduser
+from shutil import copyfile
 
 from nemo.collections.nlp.data.datasets.datasets_utils import if_exist
 
@@ -378,7 +381,7 @@ def divideData(data, infold, outfold):
     the data for three different sets"""
 
     os.makedirs(outfold, exist_ok=True)
-    shutil.copyfile(f'{infold}/ontology.json', f'{outfold}/ontology.json')
+    copyfile(f'{infold}/ontology.json', f'{outfold}/ontology.json')
 
     testListFile = []
     fin = open(f'{infold}/testListFile.json', 'r')
@@ -436,6 +439,40 @@ def divideData(data, infold, outfold):
                 train_dials.append(dialogue)
                 count_train += 1
 
+    value_dict = json.load(open(f'{outfold}/ontology.json'))
+    new_ontology = {}
+    # k = {'taxi-arrive by':list_of_values} -> {'taxi':{'arriveby': list_ofslot_values}}
+    for k, v in value_dict.items():
+        domain, slot = k.split('-')
+        slot = slot.replace(' ', '')
+        if domain in new_ontology:
+            new_ontology[domain][slot] = v
+        else:
+            new_ontology[domain] = {slot: v}
+
+    with open(f'{outfold}/value_dict.json', 'w') as f:
+        json.dump(new_ontology, f, indent=4)
+
+    # save all data base *db.json file in a db folder
+    db_fold = os.path.join(outfold, 'db')
+    os.makedirs(db_fold, exist_ok=True)
+
+    for f in os.listdir(infold):
+        if '_db.json' in f:
+            copyfile(f'{infold}/{f}', f'{db_fold}/{f}')
+        # taxi_db.json file is missing a comma in the MultiWOZ2.1 dataset
+        # check if it's so and fix
+        if f == 'taxi_db.json':
+            try:
+                with open(os.path.join(db_fold, f)) as f_:
+                    _ = json.load(f_)
+            except json.decoder.JSONDecodeError:
+                taxi_db_text = open(os.path.join(db_fold, f)).readlines()
+                taxi_db_text[2] = taxi_db_text[2].rstrip() + ',\n'
+                taxi_db_text = '{' + ''.join(taxi_db_text)[1:-2] + '}\n'
+                taxi_db_text = taxi_db_text.replace("'", '"')
+                open(os.path.join(db_fold, f), 'w').write(taxi_db_text)
+
     # save all dialogues
     with open(f'{outfold}/dev_dials.json', 'w') as f:
         json.dump(val_dials, f, indent=4)
@@ -446,7 +483,9 @@ def divideData(data, infold, outfold):
     with open(f'{outfold}/train_dials.json', 'w') as f:
         json.dump(train_dials, f, indent=4)
 
-    print(f"Saving done. Generated dialogs: {count_train} train, {count_val} val, {count_test} test.")
+    print(
+        f"Processing done and saved in `{outfold}`. Generated dialogs: {count_train} train, {count_val} val, {count_test} test."
+    )
 
 
 if __name__ == "__main__":
@@ -456,14 +495,22 @@ def divideData(data, infold, outfold):
         "--source_data_dir", required=True, type=str, help='The path to the folder containing the MultiWOZ data files.'
     )
     parser.add_argument("--target_data_dir", default='multiwoz2.1/', type=str)
+    parser.add_argument("--overwrite_files", action="store_true", help="Whether to overwrite preprocessed file")
     args = parser.parse_args()
 
-    if not exists(args.source_data_dir):
-        raise FileNotFoundError(f"{args.source_data_dir} does not exist.")
+    # Get the absolute path.
+    abs_source_data_dir = expanduser(args.source_data_dir)
+    abs_target_data_dir = expanduser(args.target_data_dir)
+
+    if not exists(abs_source_data_dir):
+        raise FileNotFoundError(f"{abs_source_data_dir} does not exist.")
 
     # Check if the files exist
-    if if_exist(args.target_data_dir, ['ontology.json', 'dev_dials.json', 'test_dials.json', 'train_dials.json']):
-        print(f'Data is already processed and stored at {args.source_data_dir}, skipping pre-processing.')
+    if (
+        if_exist(abs_target_data_dir, ['ontology.json', 'dev_dials.json', 'test_dials.json', 'train_dials.json', 'db'])
+        and not args.overwrite_files
+    ):
+        print(f'Data is already processed and stored at {abs_target_data_dir}, skipping pre-processing.')
         exit(0)
 
     fin = open('multiwoz_mapping.pair', 'r')
@@ -474,6 +521,6 @@ def divideData(data, infold, outfold):
 
     print('Creating dialogues...')
     # Process MultiWOZ dataset
-    delex_data = createData(args.source_data_dir)
+    delex_data = createData(abs_source_data_dir)
     # Divide data
-    divideData(delex_data, args.source_data_dir, args.target_data_dir)
+    divideData(delex_data, abs_source_data_dir, abs_target_data_dir)

examples/nlp/dialogue_state_tracking/dialogue_state_tracking_trade.py

@@ -30,13 +30,12 @@
 
 import nemo.core as nemo_core
 from nemo import logging
-from nemo.backends.pytorch.common import EncoderRNN
 from nemo.backends.pytorch.common.losses import CrossEntropyLossNM, LossAggregatorNM
 from nemo.collections.nlp.callbacks.state_tracking_trade_callback import eval_epochs_done_callback, eval_iter_callback
 from nemo.collections.nlp.data.datasets.multiwoz_dataset import MultiWOZDataDesc
 from nemo.collections.nlp.nm.data_layers import MultiWOZDataLayer
 from nemo.collections.nlp.nm.losses import MaskedLogLoss
-from nemo.collections.nlp.nm.trainables import TRADEGenerator
+from nemo.collections.nlp.nm.trainables import EncoderRNN, TRADEGenerator
 from nemo.utils.lr_policies import get_lr_policy
 
 parser = argparse.ArgumentParser(description='Dialogue state tracking with TRADE model on MultiWOZ dataset')
@@ -155,8 +154,8 @@ def create_pipeline(num_samples, batch_size, num_gpus, input_dropout, data_prefi
     point_outputs, gate_outputs = decoder(
         encoder_hidden=hidden,
         encoder_outputs=outputs,
-        input_lens=input_data.src_lens,
-        src_ids=input_data.src_ids,
+        dialog_lens=input_data.src_lens,
+        dialog_ids=input_data.src_ids,
         targets=input_data.tgt_ids,
     )