Truncate golden CosineScores to 6 decimals to reduce noisy diffs. (#4718

) Reduces CosineScores to 6 decimals to reduce the diffs that are generated when updating the integration_test goldens. This is to reduce noise in reviews and should not fundamentally affect the actual tests.
datacommonsorg · Nov 7, 2024 · 9d1d4e0 · 9d1d4e0
1 parent ddd3e22
commit 9d1d4e0
Show file tree

Hide file tree

Showing 8 changed files with 341 additions and 326 deletions.
diff --git a/server/integration_tests/explore_test.py b/server/integration_tests/explore_test.py
@@ -119,13 +119,23 @@ def handle_response(self,
     # TODO: Proper fix should be to make NL server more deterministic
     if 'variables' in resp:
       resp_var_to_score = {}
+      dbg['sv_matching']['CosineScore'] = _format_scores(
+          dbg['sv_matching']['CosineScore'])
       for i, sv in enumerate(dbg['sv_matching']['SV']):
-        score = dbg['sv_matching']['CosineScore'][i]
-        resp_var_to_score[sv] = float("{:.6f}".format(score))
+        resp_var_to_score[sv] = dbg['sv_matching']['CosineScore'][i]
       sorted_variables = sorted(resp['variables'],
                                 key=lambda x: (-resp_var_to_score.get(x, 0), x))
       resp['variables'] = sorted_variables
 
+    # Truncate CosineScores to 6 decimals to reduce noisy diffs.
+    for candidate in dbg['sv_matching']['MultiSV'].get('Candidates', []):
+      for part in candidate.get('Parts', []):
+        if multisv_scores := part.get('CosineScore', []):
+          part['CosineScore'] = _format_scores(multisv_scores)
+
+    if props_scores := dbg['props_matching'].get('CosineScore', []):
+      dbg['props_matching']['CosineScore'] = _format_scores(props_scores)
+
     resp['debug'] = {}
     resp['context'] = {}
     for category in resp.get('config', {}).get('categories', []):
@@ -793,3 +803,8 @@ def _del_field(d: dict, path: str):
         del tmp[p]
       else:
         tmp = tmp[p]
+
+
+# Helper function to consistently format float scores.
+def _format_scores(scores):
+  return [float("{:.6f}".format(score)) for score in scores]
diff --git a/server/integration_tests/test_data/detection_api_bio/whatisthephylumofvolvox/debug_info.json b/server/integration_tests/test_data/detection_api_bio/whatisthephylumofvolvox/debug_info.json
@@ -10,29 +10,29 @@
   "query_with_places_removed": "what is the phylum of",
   "sv_matching": {
     "CosineScore": [
-      0.35510897636413574,
-      0.3460870683193207,
-      0.3452577292919159,
-      0.34484589099884033,
-      0.3411366045475006,
-      0.3384982645511627,
-      0.3349902629852295,
-      0.3316609263420105,
-      0.32600605487823486,
-      0.3242444694042206,
-      0.32060620188713074,
-      0.31426557898521423,
-      0.3127654492855072,
-      0.3126293420791626,
-      0.31250452995300293,
-      0.31177404522895813,
-      0.3103388845920563,
-      0.3100748658180237,
-      0.3098835051059723,
-      0.3080541491508484,
-      0.3074587285518646,
-      0.3071962296962738,
-      0.30394214391708374
+      0.355109,
+      0.346087,
+      0.345258,
+      0.344846,
+      0.341137,
+      0.338498,
+      0.33499,
+      0.331661,
+      0.326006,
+      0.324244,
+      0.320606,
+      0.314266,
+      0.312765,
+      0.312629,
+      0.312505,
+      0.311774,
+      0.310339,
+      0.310075,
+      0.309883,
+      0.308054,
+      0.307459,
+      0.307196,
+      0.303942
     ],
     "MultiSV": {},
     "Query": "what is the phylum of",
@@ -64,35 +64,35 @@
   },
   "props_matching": {
     "CosineScore": [
-      1.0000001192092896,
-      0.47528401017189026,
-      0.3615417182445526,
-      0.34543079137802124,
-      0.34543079137802124,
-      0.3420097231864929,
-      0.3293308615684509,
-      0.318451464176178,
-      0.31466978788375854,
-      0.308395117521286,
-      0.30794060230255127,
-      0.3046301007270813,
-      0.2827085852622986,
-      0.281875342130661,
-      0.28082770109176636,
-      0.2804528772830963,
-      0.2742983400821686,
-      0.2715050280094147,
-      0.27071818709373474,
-      0.2695467472076416,
-      0.25990068912506104,
-      0.25990068912506104,
-      0.24827904999256134,
-      0.24662022292613983,
-      0.24632100760936737,
-      0.2461635172367096,
-      0.24212227761745453,
-      0.23135530948638916,
-      0.2296244353055954
+      1.0,
+      0.475284,
+      0.361542,
+      0.345431,
+      0.345431,
+      0.34201,
+      0.329331,
+      0.318452,
+      0.31467,
+      0.308395,
+      0.307941,
+      0.30463,
+      0.282709,
+      0.281875,
+      0.280828,
+      0.280453,
+      0.274298,
+      0.271505,
+      0.270718,
+      0.269547,
+      0.259901,
+      0.259901,
+      0.248279,
+      0.24662,
+      0.246321,
+      0.246163,
+      0.242122,
+      0.231355,
+      0.229624
     ],
     "PROP": [
       "phylum",

diff --git a/...n_tests/test_data/detection_api_bio/whattypesofgenesarefgfr1,apoe,andache/debug_info.json b/...n_tests/test_data/detection_api_bio/whattypesofgenesarefgfr1,apoe,andache/debug_info.json
@@ -12,30 +12,30 @@
   "query_with_places_removed": "what types of genes are and",
   "sv_matching": {
     "CosineScore": [
-      0.5237791538238525,
-      0.4451894760131836,
-      0.4198145866394043,
-      0.41922125220298767,
-      0.41366666555404663,
-      0.4096527099609375,
-      0.4066110849380493,
-      0.4013715386390686,
-      0.4010191559791565,
-      0.40063756704330444,
-      0.39874646067619324,
-      0.39836058020591736,
-      0.39052361249923706,
-      0.38903331756591797,
-      0.3884609639644623,
-      0.3883824050426483,
-      0.3873301148414612,
-      0.38642188906669617,
-      0.3858962655067444,
-      0.3856789469718933,
-      0.3854343295097351,
-      0.38533878326416016,
-      0.38184264302253723,
-      0.37970495223999023
+      0.523779,
+      0.445189,
+      0.419815,
+      0.419221,
+      0.413667,
+      0.409652,
+      0.406611,
+      0.401371,
+      0.401019,
+      0.400638,
+      0.398746,
+      0.398361,
+      0.390524,
+      0.389033,
+      0.388461,
+      0.388382,
+      0.38733,
+      0.386422,
+      0.385896,
+      0.385679,
+      0.385434,
+      0.385339,
+      0.381843,
+      0.379705
     ],
     "MultiSV": {},
     "Query": "what types of genes are and",
@@ -68,32 +68,32 @@
   },
   "props_matching": {
     "CosineScore": [
-      0.9086110591888428,
-      0.7445294260978699,
-      0.7330226898193359,
-      0.7175887823104858,
-      0.6834690570831299,
-      0.6590196490287781,
-      0.6357895731925964,
-      0.6340402364730835,
-      0.6171568036079407,
-      0.5988123416900635,
-      0.5958784818649292,
-      0.5850145220756531,
-      0.5832706689834595,
-      0.5810792446136475,
-      0.576904296875,
-      0.5432419776916504,
-      0.5399729013442993,
-      0.5320467948913574,
-      0.5309585332870483,
-      0.5142703056335449,
-      0.5125279426574707,
-      0.46105992794036865,
-      0.4441080689430237,
-      0.39822953939437866,
-      0.3944427967071533,
-      0.39229002594947815
+      0.908611,
+      0.744529,
+      0.733023,
+      0.717589,
+      0.683469,
+      0.65902,
+      0.63579,
+      0.63404,
+      0.617157,
+      0.598812,
+      0.595879,
+      0.585015,
+      0.583271,
+      0.581079,
+      0.576904,
+      0.543242,
+      0.539973,
+      0.532047,
+      0.530959,
+      0.51427,
+      0.512528,
+      0.46106,
+      0.444108,
+      0.398229,
+      0.394443,
+      0.39229
     ],
     "PROP": [
       "typeOfGene",