narrow feature space for full study

asreview · Jan 23, 2025 · 3e7476c · 3e7476c
1 parent e722d81
commit 3e7476c
Show file tree

Hide file tree

Showing 3 changed files with 8 additions and 8 deletions.
diff --git a/asreview2-optuna/classifiers.py b/asreview2-optuna/classifiers.py
@@ -10,7 +10,7 @@
 
 def naive_bayes_params(trial: optuna.trial.FrozenTrial):
     # Use logarithmic normal distribution for alpha (alpha effect is non-linear)
-    alpha = trial.suggest_float("alpha", 0.1, 100, log=True)
+    alpha = trial.suggest_float("alpha", 0.5, 50, log=True)
     #alpha = trial.suggest_float("nb__alpha", 1.0, 15.0)
     return {"alpha": alpha}
 

diff --git a/asreview2-optuna/feature_extractors.py b/asreview2-optuna/feature_extractors.py
@@ -7,14 +7,14 @@
 def tfidf_params(trial: optuna.trial.FrozenTrial):
     #max_features = trial.suggest_int("tfidf__max_features", 15_000, 50_000)
 
-    max_df = trial.suggest_float("tfidf__max_df", 0.5, 1.0)
+    max_df = trial.suggest_float("tfidf__max_df", 0.6, 0.9)
 
-    min_df = trial.suggest_int("tfidf__min_df", 1, 10)
+    min_df = trial.suggest_int("tfidf__min_df", 1, 5)
 
-    max_ngram_range = trial.suggest_int("tfidf__max_ngram_range", 1, 3)
-    ngram_range = (1, max_ngram_range)
+    #max_ngram_range = trial.suggest_int("tfidf__max_ngram_range", 1, 3)
+    ngram_range = (1, 2)
 
-    sublinear_tf = trial.suggest_categorical("tfidf__sublinear_tf", [True, False])
+    sublinear_tf = True#trial.suggest_categorical("tfidf__sublinear_tf", [True, False])
 
     return {
         #"max_features": max_features,

diff --git a/asreview2-optuna/main.py b/asreview2-optuna/main.py
@@ -20,7 +20,7 @@
 
 # Study variables
 VERSION = 1
-STUDY_SET = "demo"
+STUDY_SET = "full"
 PICKLE_FOLDER_PATH = Path("synergy-dataset", "pickles")
 CLASSIFIER_TYPE = "nb"  # Options: "nb", "log", "svm", "rf"
 FEATURE_EXTRACTOR_TYPE = "tfidf"  # Options: "tfidf", "onehot"
@@ -162,7 +162,7 @@ def process_row(row, clf_params, fe_params, ratio):
 def objective_report(report_order):
     def objective(trial):
         # Use normal distribution for ratio (ratio effect is linear)
-        ratio = trial.suggest_float("ratio", 1.0, 5.0)
+        ratio = trial.suggest_float("ratio", 1.0, 3.0)
         # ratio = 1.5
         clf_params = classifier_params[CLASSIFIER_TYPE](trial)
         fe_params = feature_extractor_params[FEATURE_EXTRACTOR_TYPE](trial)