Refactor FeatureType (#462)

* fix bugs. * add test * move utils.serialization
neulab · Sep 9, 2022 · df48ff1 · df48ff1
1 parent 7c1bbce
commit df48ff1
Show file tree

Hide file tree

Showing 29 changed files with 625 additions and 170 deletions.
diff --git a/data/system_outputs/fb15k-237/test-kg-prediction-user-defined.json b/data/system_outputs/fb15k-237/test-kg-prediction-user-defined.json
@@ -2,6 +2,7 @@
   "metadata": {
     "custom_features": {
       "rel_type": {
+        "cls_name": "Value",
         "dtype": "string",
         "description": "symmetric or asymmetric",
         "num_buckets": 2
@@ -160,4 +161,4 @@
       "example_id": "10"
     }
   ]
-}
+}
diff --git a/data/system_outputs/fb15k-237/test-user-defined-features-prediction-short.json b/data/system_outputs/fb15k-237/test-user-defined-features-prediction-short.json
@@ -2,11 +2,13 @@
   "metadata": {
     "custom_features": {
       "toy_string_feature": {
+        "cls_name": "Value",
         "dtype": "string",
         "description": "most specific (highest) entity type level of true tail entity",
         "num_buckets": 8
       },
       "toy_float_feature": {
+        "cls_name": "Value",
         "dtype": "float",
         "description": "just a toy float feature for testing",
         "num_buckets": 4
@@ -1215,4 +1217,4 @@
       "example_id": "/m/099c8n\t/award/award_category/nominees./award/award_nomination/nominated_for\t/m/064lsn"
     }
   ]
-}
+}
diff --git a/data/system_outputs/fb15k-237/user-defined.json b/data/system_outputs/fb15k-237/user-defined.json
@@ -2,6 +2,7 @@
   "metadata": {
     "custom_features": {
       "rel_type": {
+        "cls_name": "Value",
         "dtype": "string",
         "description": "symmetric or asymmetric",
         "num_buckets": 2
@@ -160,4 +161,4 @@
       "example_id": "10"
     }
   ]
-}
+}
diff --git a/data/system_outputs/sst2_tabclass/sst2-tabclass-dataset.json b/data/system_outputs/sst2_tabclass/sst2-tabclass-dataset.json
@@ -3,6 +3,7 @@
     "custom_features": {
       "example": {
         "total_words": {
+          "cls_name": "Value",
           "dtype": "float",
           "description": "total number of words"
         }

diff --git a/data/system_outputs/sst2_tabreg/sst2-tabreg-dataset.json b/data/system_outputs/sst2_tabreg/sst2-tabreg-dataset.json
@@ -3,6 +3,7 @@
     "custom_features": {
       "example": {
         "total_words": {
+          "cls_name": "Value",
           "dtype": "float",
           "description": "total number of words"
         }

diff --git a/explainaboard/analysis/analyses.py b/explainaboard/analysis/analyses.py
@@ -8,12 +8,12 @@
 
 import explainaboard.analysis.bucketing
 from explainaboard.analysis.case import AnalysisCase, AnalysisCaseCollection
-from explainaboard.analysis.feature import FeatureType
+from explainaboard.analysis.feature import FeatureType, get_feature_type_serializer
 from explainaboard.analysis.performance import BucketPerformance, Performance
 from explainaboard.metrics.metric import Metric, MetricConfig, MetricStats
 from explainaboard.metrics.registry import metric_config_from_dict
 from explainaboard.utils.logging import get_logger
-from explainaboard.utils.typing_utils import unwrap, unwrap_generator
+from explainaboard.utils.typing_utils import narrow, unwrap, unwrap_generator
 
 
 @dataclass
@@ -325,7 +325,13 @@ class AnalysisLevel:
 
     @staticmethod
     def from_dict(dikt: dict):
-        features = {k: FeatureType.from_dict(v) for k, v in dikt['features'].items()}
+        ft_serializer = get_feature_type_serializer()
+
+        features = {
+            # See https://github.com/python/mypy/issues/4717
+            k: narrow(FeatureType, ft_serializer.deserialize(v))  # type: ignore
+            for k, v in dikt['features'].items()
+        }
         metric_configs = [metric_config_from_dict(v) for v in dikt['metric_configs']]
         return AnalysisLevel(
             name=dikt['name'],