fix(app): avoid KeyError when no filter OP provides tunable stats

app.py

@@ -198,6 +198,19 @@ def get_min_max_step(data):
     return min_value, max_value, step
 
 
+def combine_conds(conds, num_samples):
+    """Combine the per-OP boolean conditions into a single mask over samples.
+
+    ``conds`` is empty whenever none of the OPs in the config is covered by
+    ``op_stats_dict``. ``np.all`` over an empty list collapses to a scalar,
+    which ``DataFrame.loc`` then reads as a label instead of a boolean mask and
+    raises ``KeyError: True``, so keep all the samples explicitly in that case.
+    """
+    if not conds:
+        return np.ones(num_samples, dtype=bool)
+    return np.all([list(cond.values())[0] for cond in conds], axis=0)
+
+
 op_stats_dict = {
     "alphanumeric_filter": [StatsKeys.alpha_token_ratio, StatsKeys.alnum_ratio],
     "average_line_length_filter": [StatsKeys.avg_line_length],
@@ -336,7 +349,12 @@ def set_sliders(total_stats, ordered):
         if ordered:
             all_conds = [True if i in filtered_stats.index else False for i in range(len(stats))]
         else:
-            all_conds = np.all([list(cond.values())[0] for cond in conds], axis=0)
+            if not conds:
+                st.warning(
+                    "None of the OPs in this config provides stats that this demo can tune, "
+                    "so no sample is filtered out below."
+                )
+            all_conds = combine_conds(conds, len(stats))
         ds = pd.DataFrame(dataset)
         Visualize.display_dataset(ds, all_conds, show_num, "Retained samples", "docs")
         st.download_button(

tests/tools/test_app.py

@@ -0,0 +1,44 @@
+import importlib.util
+import unittest
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+from data_juicer.utils.unittest_utils import DataJuicerTestCaseBase
+
+ROOT = Path(__file__).resolve().parents[2]
+
+
+class AppTest(DataJuicerTestCaseBase):
+    def setUp(self):
+        super().setUp()
+        spec = importlib.util.spec_from_file_location("dj_app", ROOT / "app.py")
+        self.app = importlib.util.module_from_spec(spec)
+        spec.loader.exec_module(self.app)
+
+    def test_combine_conds_without_cond(self):
+        # None of the OPs in the config is covered by op_stats_dict, so no
+        # condition is collected. The combined mask must still cover every
+        # sample: a scalar would be read as a label by DataFrame.loc and raise
+        # KeyError instead of selecting rows.
+        dataframe = pd.DataFrame({"text": ["a", "b", "c"]})
+        all_conds = self.app.combine_conds([], len(dataframe))
+        self.assertEqual(len(all_conds), len(dataframe))
+        self.assertTrue(np.all(all_conds))
+        self.assertEqual(len(dataframe.loc[all_conds]), 3)
+        self.assertEqual(len(dataframe.loc[np.invert(all_conds)]), 0)
+
+    def test_combine_conds_with_conds(self):
+        dataframe = pd.DataFrame({"text": ["a", "b", "c"]})
+        conds = [
+            {("1 text_length_filter", "text_len"): [True, True, False]},
+            {("2 words_num_filter", "num_words"): [True, False, True]},
+        ]
+        all_conds = self.app.combine_conds(conds, len(dataframe))
+        self.assertEqual(list(all_conds), [True, False, False])
+        self.assertEqual(len(dataframe.loc[all_conds]), 1)
+
+
+if __name__ == "__main__":
+    unittest.main()