SGD DST (#612)

* sgd baseline added Signed-off-by: Evelina Bakhturina <[email protected]> * sgd baseline added Signed-off-by: Evelina Bakhturina <[email protected]> * intent labels switch to ind, masking moved to model Signed-off-by: Evelina Bakhturina <[email protected]> * headers added, clean up, input_ex moved Signed-off-by: Evelina Bakhturina <[email protected]> * pin_memory added Signed-off-by: Evelina Bakhturina <[email protected]> * pin_memory Signed-off-by: Evelina Bakhturina <[email protected]> * multiwoz2.1 support and sgd data augmentation Signed-off-by: Yang Zhang <[email protected]> * changed loss to use sum instead of aver, added docstrings and comments, clean up, Signed-off-by: Evelina Bakhturina <[email protected]> * data aug renamed + header Signed-off-by: Evelina Bakhturina <[email protected]> * loss normalization for the batch_size added Signed-off-by: Evelina Bakhturina <[email protected]> * more docstring and loss reduction Signed-off-by: Evelina Bakhturina <[email protected]> * lgtm fixes Signed-off-by: Evelina Bakhturina <[email protected]> * fix lgtm Signed-off-by: Yang Zhang <[email protected]> * jenkins test for sgd Signed-off-by: Evelina Bakhturina <[email protected]> * rebase Signed-off-by: Evelina Bakhturina <[email protected]> * jenkins path Signed-off-by: Evelina Bakhturina <[email protected]> * jenkins path Signed-off-by: Evelina Bakhturina <[email protected]> * Fixed loss function to guard non-active losses. Signed-off-by: Vahid Noroozi <[email protected]> * Fixed crossenropy loss module to handle empty logits and labels. Signed-off-by: Vahid Noroozi <[email protected]> * lgtm fixes Signed-off-by: Evelina Bakhturina <[email protected]> * removed textdatalayer dep Signed-off-by: Evelina Bakhturina <[email protected]> * fix num2string for augmentation Signed-off-by: Yang Zhang <[email protected]> * fix lgtm Signed-off-by: Yang Zhang <[email protected]> * adding inflect to nlp requirements Signed-off-by: Yang Zhang <[email protected]> * added attention head transform Signed-off-by: Yang Zhang <[email protected]> * masks moved to the dataset Signed-off-by: Evelina Bakhturina <[email protected]> * merge Signed-off-by: Evelina Bakhturina <[email protected]> * headers updated Signed-off-by: Evelina Bakhturina <[email protected]> * remove unused import Signed-off-by: Evelina Bakhturina <[email protected]> Co-authored-by: Yang Zhang <[email protected]> Co-authored-by: Yang Zhang <[email protected]> Co-authored-by: Vahid Noroozi <[email protected]>
NVIDIA · May 22, 2020 · 09412e2 · 09412e2
1 parent 4951396
commit 09412e2
Show file tree

Hide file tree

Showing 33 changed files with 6,133 additions and 6 deletions.
diff --git a/Jenkinsfile b/Jenkinsfile
@@ -193,11 +193,22 @@ pipeline {
         }
         stage ('Punctuation and Classification Training/Inference Test') {
           steps {
-            sh 'cd examples/nlp/token_classification && CUDA_VISIBLE_DEVICES=1 python punctuation_capitalization.py --data_dir /home/TestData/nlp/token_classification_punctuation/ --work_dir punctuation_output --save_epoch_freq 1 --num_epochs 1 --save_step_freq -1 --batch_size 2'
+            sh 'cd examples/nlp/token_classification && CUDA_VISIBLE_DEVICES=1 python punctuation_capitalization.py \
+            --data_dir /home/TestData/nlp/token_classification_punctuation/ --work_dir punctuation_output --save_epoch_freq 1 \
+            --num_epochs 1 --save_step_freq -1 --batch_size 2'
             sh 'cd examples/nlp/token_classification && DATE_F=$(ls punctuation_output/) && DATA_DIR="/home/TestData/nlp/token_classification_punctuation" && CUDA_VISIBLE_DEVICES=1 python punctuation_capitalization_infer.py --checkpoint_dir punctuation_output/$DATE_F/checkpoints/ --punct_labels_dict $DATA_DIR/punct_label_ids.csv --capit_labels_dict $DATA_DIR/capit_label_ids.csv'
             sh 'rm -rf examples/nlp/token_classification/punctuation_output'
           }
         }
+        stage('SGD Test') {
+          steps {
+            sh 'cd examples/nlp/dialogue_state_tracking && CUDA_VISIBLE_DEVICES=0 python dialogue_state_tracking_sgd.py \
+            --data_dir /home/TestData/nlp/sgd/ --schema_embedding_dir /home/TestData/nlp/sgd/embeddings/ --eval_dataset dev \
+            --dialogues_example_dir /home/TestData/nlp/sgd/dialogue_example_dir/ --work_dir sgd_output --task DEBUG \
+            --num_epochs 1 --save_epoch_freq=0'
+            sh 'rm -rf examples/nlp/dialogue_state_tracking/sgd_output'
+          }
+        }
       }
     }
 

diff --git a/examples/nlp/dialogue_state_tracking/data/dialogue_augmentation_for_sgd_format.py b/examples/nlp/dialogue_state_tracking/data/dialogue_augmentation_for_sgd_format.py
diff --git a/examples/nlp/dialogue_state_tracking/data/multiwoz/__init__.py b/examples/nlp/dialogue_state_tracking/data/multiwoz/__init__.py
@@ -0,0 +1,16 @@
+# =============================================================================
+# Copyright 2020 NVIDIA. All Rights Reserved.
+# Copyright 2019 The Google Research Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# =============================================================================
diff --git a/examples/nlp/dialogue_state_tracking/data/multiwoz/correct_categorical_state_values.tsv b/examples/nlp/dialogue_state_tracking/data/multiwoz/correct_categorical_state_values.tsv
@@ -0,0 +1,18 @@
+alpha-milton	alpha milton
+any	dontcare
+bed and breakfast	guesthouse
+boating	boat
+cam	cambridge
+concert	concerthall
+concert hall	concerthall
+guest house	guesthouse
+guesthouses	guesthouse
+moderate|cheap	cheap|moderate
+museum kettles yard	museum
+mutiple sports	multiple sports
+nightclub	night club
+acorn guesthouse	acorn guest house
+swimmingpool	swimming pool
+sports	multiple sports
+pool	swimming pool
+theater	theatre