more settings

app.py

@@ -15,6 +15,7 @@ COLORS = ['rgb(252, 132, 100)','rgb(252, 148, 116)','rgb(252, 166, 137)','rgb(25
 doc2query = Doc2Query(MODEL, append=True, num_samples=5)
 electra = ElectraScorer()
 query_scorer = QueryScorer(electra)
+query_filter = QueryFilter(p=0.5, append=False)
 
 COLAB_NAME = 'pyterrier_doc2query.ipynb'
 COLAB_INSTALL = '''
@@ -75,13 +76,28 @@ def generate_vis(df):
 ''')
   return '\n'.join(result)
 
-def predict_mm(input, model, num_samples, score_model):
+def predict_mm(input, model, num_samples, score_model, filter_pct):
   assert model == MODEL
   assert score_model == SCORE_MODEL
   doc2query.append = False
   doc2query.num_samples = num_samples
-  pipeline = doc2query >> query_scorer
-  code = f'''import pyterrier as pt ; pt.init()
+  if filter_pct > 0:
+    query_filter.t = PERCENTILES_BY_5[filter_pct//5-1]
+    pipeline = doc2query >> query_scorer >> query_filter
+    code = f'''import pyterrier as pt ; pt.init()
+import pandas as pd
+from pyterrier_doc2query import Doc2Query, QueryScorer, QueryFilter
+from pyterrier_dr import ElectraScorer
+
+doc2query = Doc2Query({repr(model)}, append=False, num_samples={num_samples})
+scorer = ElectraScorer({repr(score_model)})
+pipeline = doc2query >> QueryScorer(scorer) >> QueryFilter(t={query_filter.t})
+
+pipeline({df2code(input)})
+'''
+  else:
+    pipeline = doc2query >> query_scorer
+    code = f'''import pyterrier as pt ; pt.init()
 import pandas as pd
 from pyterrier_doc2query import Doc2Query, QueryScorer
 from pyterrier_dr import ElectraScorer
@@ -138,8 +154,14 @@ interface(
     ), gr.Dropdown(
       choices=[SCORE_MODEL],
       value=SCORE_MODEL,
-      label='Filter',
+      label='Scorer',
       interactive=False,
+    ), gr.Slider(
+      minimum=0,
+      maximum=95,
+      value=10,
+      step=5,
+      label='Filter (top % of queries)'
     )],
   ),
   MarkdownFile('wrapup.md'),