accept a parameter for different ways to resolve human ambiguity

.github/workflows/CICD-dev.yml

@@ -62,7 +62,7 @@ jobs:
   refresh-dev-staging-deployment:
     name: Refresh Dev Staging Deployment
     needs: build-and-push-dev-docker-image
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-22.04
     steps:
     - name: Refresh Dev inatvisionapi
       uses: actions/github-script@v6
@@ -89,7 +89,7 @@ jobs:
     name: Notify Slack
     needs: build-and-push-dev-docker-image
     if: ${{ success() || failure() }}
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-22.04
     steps:
     - uses: iRoachie/slack-github-actions@v2.3.2
       if: env.SLACK_WEBHOOK_URL != null

.github/workflows/CICD-main.yml

@@ -65,7 +65,7 @@ jobs:
   refresh-main-staging-deployment:
     name: Refresh Main Staging Deployment
     needs: build-and-push-main-docker-image
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-22.04
     steps:
     - name: Refresh Main inatvisionapi
       uses: actions/github-script@v6
@@ -92,7 +92,7 @@ jobs:
     name: Notify Slack
     needs: build-and-push-main-docker-image
     if: ${{ success() || failure() }}
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-22.04
     steps:
     - uses: iRoachie/slack-github-actions@v2.3.2
       if: env.SLACK_WEBHOOK_URL != null

lib/inat_inferrer.py

@@ -613,7 +613,10 @@ def common_ancestor_from_aggregated_scores(
 
         return common_ancestor
 
-    def limit_leaf_scores_that_include_humans(self, leaf_scores):
+    def limit_leaf_scores_that_include_humans(self, leaf_scores, strategy=None):
+        if strategy == "never_exclude":
+            return leaf_scores
+
         if self.taxonomy.human_taxon is None:
             return leaf_scores
 
@@ -636,6 +639,19 @@ def limit_leaf_scores_that_include_humans(self, leaf_scores):
             if human_score_margin > 1.5:
                 return top_results.head(1)
 
+        # if requesting a more limited approach to human exclusion
+        if strategy == "limited" and self.taxonomy.mammals_taxon is not None:
+            mammals_results = top_results.query(
+                f"left > {self.taxonomy.mammals_taxon['left']} and "
+                f"right < {self.taxonomy.mammals_taxon['right']}"
+            )
+            # if there is only 1 Mammals taxon (human), return all non-human results
+            if mammals_results.index.size == 1:
+                non_human_results = top_results.query(
+                    f"taxon_id != {self.taxonomy.human_taxon['taxon_id']}"
+                )
+                return non_human_results
+
         # otherwise return no results
         return leaf_scores.head(0)
 

lib/inat_vision_api.py

@@ -178,7 +178,8 @@ def score_image(self, form, file_path, lat, lng, filter_taxon_id, geomodel,
             embedding = predictions_for_image["features"]
             return InatVisionAPIResponses.aggregated_object_response(
                 leaf_scores, aggregated_scores, self.inferrer,
-                embedding=embedding
+                embedding=embedding,
+                human_exclusion_strategy=form.human_exclusion.data
             )
 
         # legacy dict response
@@ -192,6 +193,7 @@ def score_image(self, form, file_path, lat, lng, filter_taxon_id, geomodel,
                 self.inferrer,
                 common_ancestor_rank_type=common_ancestor_rank_type,
                 embedding=embedding,
+                human_exclusion_strategy=form.human_exclusion.data,
                 debug=self.debug
             )
 

lib/inat_vision_api_responses.py

@@ -21,10 +21,13 @@ def array_response(leaf_scores, inferrer):
 
     @staticmethod
     def object_response(leaf_scores, inferrer, common_ancestor_rank_type=None,
-                        embedding=None, debug=False):
+                        embedding=None, debug=False, human_exclusion_strategy=None):
         leaf_scores = InatVisionAPIResponses.limit_leaf_scores_for_response(leaf_scores)
         leaf_scores = InatVisionAPIResponses.update_leaf_scores_scaling(leaf_scores)
-        post_human_exclusion_scores = inferrer.limit_leaf_scores_that_include_humans(leaf_scores)
+
+        post_human_exclusion_scores = inferrer.limit_leaf_scores_that_include_humans(
+            leaf_scores, strategy=human_exclusion_strategy
+        )
         human_exclusion_cleared_results = False
         if not leaf_scores.empty and post_human_exclusion_scores.empty:
             human_exclusion_cleared_results = True
@@ -85,7 +88,9 @@ def aggregated_tree_response(aggregated_scores, inferrer):
         return "<pre>" + "<br/>".join(printable_tree) + "</pre>"
 
     @staticmethod
-    def aggregated_object_response(leaf_scores, aggregated_scores, inferrer, embedding=None):
+    def aggregated_object_response(
+        leaf_scores, aggregated_scores, inferrer, embedding=None, human_exclusion_strategy=None
+    ):
         top_leaf_combined_score = aggregated_scores.query(
             "leaf_class_id.notnull()"
         ).sort_values(
@@ -100,7 +105,10 @@ def aggregated_object_response(leaf_scores, aggregated_scores, inferrer, embeddi
             "normalized_aggregated_combined_score",
             ascending=False
         ).head(100)
-        top_100_leaves = inferrer.limit_leaf_scores_that_include_humans(top_100_leaves)
+
+        top_100_leaves = inferrer.limit_leaf_scores_that_include_humans(
+            top_100_leaves, strategy=human_exclusion_strategy
+        )
 
         aggregated_scores = InatVisionAPIResponses.update_aggregated_scores_scaling(
             aggregated_scores

lib/model_taxonomy_dataframe.py

@@ -7,6 +7,7 @@ class ModelTaxonomyDataframe:
     def __init__(self, path, thresholds_path):
         self.load_mapping(path, thresholds_path)
         self.set_human_taxon()
+        self.set_mammals_taxon()
 
     def load_mapping(self, path, thresholds_path):
         self.df = pd.read_csv(
@@ -78,6 +79,20 @@ def set_human_taxon(self):
 
         self.human_taxon = human_rows.iloc[0]
 
+    def set_mammals_taxon(self):
+        self.mammals_taxon = None
+        if self.human_taxon is None:
+            return
+
+        mammals_rows = self.df.query(
+            f"name == 'Mammalia' and left < {self.human_taxon['left']} and "
+            f"right > {self.human_taxon['right']}"
+        )
+        if mammals_rows.empty:
+            return
+
+        self.mammals_taxon = mammals_rows.iloc[0]
+
     @staticmethod
     def children(df, taxon_id):
         if taxon_id == 0:

lib/templates/home.html

@@ -49,6 +49,12 @@ <h2>Slim vs Legacy Model</h2>
       <option value="major">Major Ranks (genus to kingdom)</option>
     </select>
     <br/>
+    <select name="human_exclusion">
+      <option>Exclude Humans</option>
+      <option value="limited">Limited Human Exclusion</option>
+      <option value="never_exclude">Never Exclude Humans</option>
+    </select>
+    <br/>
     <br/>
     <button type="submit">Submit</button>
   </form>

lib/web_forms.py

@@ -15,4 +15,5 @@ class ImageForm(FlaskForm):
     aggregated = StringField("aggregated")
     return_embedding = StringField("return_embedding")
     common_ancestor_rank_type = StringField("common_ancestor_rank_type")
+    human_exclusion = StringField("human_exclusion")
     format = StringField("format")