[WIP] Fixing data cache (#1314)

Author: parmeet

.circleci/cached_datasets_list.txt

@@ -0,0 +1,21 @@
+IMDB
+AG_NEWS
+SogouNews
+DBpedia
+YelpReviewPolarity
+YelpReviewFull
+YahooAnswers
+AmazonReviewPolarity
+AmazonReviewFull
+UDPOS
+CoNLL2000Chunking
+Multi30k
+IWSLT2016
+IWSLT2017
+WMT14
+WikiText2
+WikiText103
+PennTreebank
+SQuAD1
+SQuAD2
+EnWik9

.circleci/config.yml

@@ -44,7 +44,9 @@ commands:
     steps:
       - run:
           name: Generate CCI cache key
-          command: echo "$(date "+%D")" > .cachekey
+          command:
+            echo "$(date "+%D")" > .cachekey
+            cat cached_datasets_list.txt >> .cachekey
       - persist_to_workspace:
           root: .
           paths:

.circleci/config.yml.in

@@ -44,7 +44,9 @@ commands:
     steps:
       - run:
           name: Generate CCI cache key
-          command: echo "$(date "+%D")" > .cachekey
+          command:
+            echo "$(date "+%D")" > .cachekey
+            cat cached_datasets_list.txt >> .cachekey
       - persist_to_workspace:
           root: .
           paths:

test/common/cache_utils.py

@@ -9,11 +9,14 @@
 def check_cache_status():
     assert os.path.exists(CACHE_STATUS_FILE), "Cache status file does not exists"
     with open(CACHE_STATUS_FILE, 'r') as f:
+        missing_datasets = []
         cache_status = json.load(f)
         for dataset_name in cache_status:
             for split in cache_status[dataset_name]:
                 if cache_status[dataset_name][split]['status'] == "fail":
-                    raise FileNotFoundError("Failing all raw dataset unit tests as cache is missing atleast one raw dataset")
+                    missing_datasets.append(dataset_name + '_' + split)
+        if missing_datasets:
+            raise FileNotFoundError("Failing all raw dataset unit tests as cache is missing {} datasets".format(missing_datasets))
 
 
 def generate_data_cache():
@@ -30,7 +33,7 @@ def generate_data_cache():
         if dataset_name not in cache_status:
             cache_status[dataset_name] = {}
         try:
-            if dataset_name == "Multi30k" or dataset_name == 'WMT14':
+            if dataset_name == 'WMT14':
                 _ = torchtext.experimental.datasets.raw.DATASETS[dataset_name](split=split)
             else:
                 _ = torchtext.datasets.DATASETS[dataset_name](split=split)