updated model selection logic

interfaces/cap.py

@@ -47,43 +47,20 @@ def check_huggingface_path(checkpoint_path: str):
 def build_huggingface_path(language: str, domain: str):
     language = language.lower()
     base_path = "xlm-roberta-large"
-
-    if language == "english" and (domain == "media" or domain == "legislative"):
-        lang_domain_path = f"poltextlab/{base_path}-{language}-{domain}-cap-v4"
-        return lang_domain_path
-    else:
-        lang_domain_path = f"poltextlab/{base_path}-{language}-{domain}-cap-v3"
-        
     lang_path = f"poltextlab/{base_path}-{language}-cap-v3"
 
-    path_map = {
-        "L": lang_path,
-        "L-D": lang_domain_path,
-        "X": lang_domain_path,
-    }
-    value = None
-
-    try:
-        lang_domain_table = pd.read_csv("language_domain_models.csv")
-        lang_domain_table["language"] = lang_domain_table["language"].str.lower()
-        lang_domain_table.columns = lang_domain_table.columns.str.lower()
-        # get the row for the language and them get the value from the domain column
-        row = lang_domain_table[(lang_domain_table["language"] == language)]
-        tmp = row.get(domain)
-        if not tmp.empty:
-            value = tmp.iloc[0]
-    except (AttributeError, FileNotFoundError):
-        value = None
-
-    if language == 'english':
-        model_path = lang_path
-    else:
-        model_path = "poltextlab/xlm-roberta-large-pooled-cap"
-
-    if check_huggingface_path(model_path):
-        return model_path
+    # some custom mapping
+    if language in ["english"]:
+        if domain in ["media", "legislative"]:
+            return f"poltextlab/{base_path}-{language}-{domain}-cap-v4"
+        elif domain in ["social"]:
+            return f"poltextlab/{base_path}-{language}-{domain}-cap-v3"
+        return lang_path
+    
     else:
-        return "poltextlab/xlm-roberta-large-pooled-cap"
+        if domain in ["social"]:
+            return f"poltextlab/{base_path}-{domain}-cap-v3"
+        return "poltextlab/xlm-roberta-large-pooled-cap-v3"
         
 
 def predict(text, model_id, tokenizer_id):