Add TASB and TCT-V2 integration test (Issue#638) (castorini#658)

* add tct_colbert-v2 integration test

* add distilbert_tasb integration test

Co-authored-by: Lin Jack <[email protected]>
Co-authored-by: justram <[email protected]>

integrations/dense/test_distilbert_tasb.py

@@ -0,0 +1,68 @@
+#
+# Pyserini: Reproducible IR research with sparse and dense representations
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+"""Integration tests for TASB model using on-the-fly query encoding."""
+
+import os
+import socket
+import unittest
+
+from integrations.utils import clean_files, run_command, parse_score
+from pyserini.dsearch import QueryEncoder
+from pyserini.search import get_topics
+
+
+class TestSearchIntegration(unittest.TestCase):
+    def setUp(self):
+        self.temp_files = []
+        self.threads = 12
+        self.batch_size = 36
+
+        # Hard-code larger values for internal servers
+        if socket.gethostname().startswith('damiano') or socket.gethostname().startswith('orca'):
+            self.threads = 36
+            self.batch_size = 144
+
+    def test_msmarco_passage_distilbert_kd_tas_b_bf_otf(self):
+        output_file = 'test_run.msmarco-passage.distilbert-dot-tas_b-b256.bf.tsv'
+        self.temp_files.append(output_file)
+        cmd1 = f'python -m pyserini.dsearch --topics msmarco-passage-dev-subset \
+                             --index msmarco-passage-distilbert-dot-tas_b-b256-bf \
+                             --encoder sebastian-hofstaetter/distilbert-dot-tas_b-b256-msmarco \
+                             --batch-size {self.batch_size} \
+                             --threads {self.threads} \
+                             --output {output_file} \
+                             --output-format msmarco'
+        cmd2 = f'python -m pyserini.eval.msmarco_passage_eval msmarco-passage-dev-subset {output_file}'
+        status = os.system(cmd1)
+        stdout, stderr = run_command(cmd2)
+        score = parse_score(stdout, "MRR @10")
+        self.assertEqual(status, 0)
+        self.assertEqual(stderr, '')
+        self.assertAlmostEqual(score, 0.3444, delta=0.0001)
+
+    def test_msmarco_passage_distilbert_kd_tas_b_encoded_queries(self):
+        encoder = QueryEncoder.load_encoded_queries('distilbert_tas_b-msmarco-passage-dev-subset')
+        topics = get_topics('msmarco-passage-dev-subset')
+        for t in topics:
+            self.assertTrue(topics[t]['title'] in encoder.embedding)
+
+    def tearDown(self):
+        clean_files(self.temp_files)
+
+
+if __name__ == '__main__':
+    unittest.main()

integrations/dense/test_tct_colbert-v2.py

@@ -0,0 +1,145 @@
+#
+# Pyserini: Reproducible IR research with sparse and dense representations
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+"""Integration tests for TCT-ColBERT model using on-the-fly query encoding."""
+
+import os
+import socket
+import unittest
+
+from integrations.utils import clean_files, run_command, parse_score
+from pyserini.dsearch import QueryEncoder
+from pyserini.search import get_topics
+
+
+class TestSearchIntegration(unittest.TestCase):
+    def setUp(self):
+        self.temp_files = []
+        self.threads = 12
+        self.batch_size = 36
+
+        # Hard-code larger values for internal servers
+        if socket.gethostname().startswith('damiano') or socket.gethostname().startswith('orca'):
+            self.threads = 36
+            self.batch_size = 144
+
+    def test_msmarco_passage_tct_colbert_v2_bf_otf(self):
+        output_file = 'test_run.msmarco-passage.tct_colbert-v2.bf-otf.tsv'
+        self.temp_files.append(output_file)
+        cmd1 = f'python -m pyserini.dsearch --topics msmarco-passage-dev-subset \
+                             --index msmarco-passage-tct_colbert-v2-bf \
+                             --encoder castorini/tct_colbert-v2-msmarco \
+                             --batch-size {self.batch_size} \
+                             --threads {self.threads} \
+                             --output {output_file} \
+                             --output-format msmarco'
+        cmd2 = f'python -m pyserini.eval.msmarco_passage_eval msmarco-passage-dev-subset {output_file}'
+        status = os.system(cmd1)
+        stdout, stderr = run_command(cmd2)
+        score = parse_score(stdout, "MRR @10")
+        self.assertEqual(status, 0)
+        self.assertAlmostEqual(score, 0.3439, places=4)
+
+    def test_msmarco_passage_tct_colbert_v2_hn_otf(self):
+        output_file = 'test_run.msmarco-passage.tct_colbert-v2-hn.bf-otf.tsv'
+        self.temp_files.append(output_file)
+        cmd1 = f'python -m pyserini.dsearch --topics msmarco-passage-dev-subset \
+                             --index msmarco-passage-tct_colbert-v2-hn-bf \
+                             --encoder castorini/tct_colbert-v2-hn-msmarco \
+                             --output {output_file} \
+                             --output-format msmarco '
+        cmd2 = f'python -m pyserini.eval.msmarco_passage_eval msmarco-passage-dev-subset {output_file}'
+        status = os.system(cmd1)
+        stdout, stderr = run_command(cmd2)
+        score = parse_score(stdout, "MRR @10")
+        self.assertEqual(status, 0)
+        self.assertAlmostEqual(score, 0.3542, places=4)
+
+    def test_msmarco_passage_tct_colbert_v2_hnp_otf(self):
+        output_file = 'test_run.msmarco-passage.tct_colbert-v2-hnp.bf-otf.tsv'
+        self.temp_files.append(output_file)
+        cmd1 = f'python -m pyserini.dsearch --topics msmarco-passage-dev-subset \
+                             --index msmarco-passage-tct_colbert-v2-hnp-bf \
+                             --encoder castorini/tct_colbert-v2-hnp-msmarco \
+                             --output {output_file} \
+                             --output-format msmarco '
+        cmd2 = f'python -m pyserini.eval.msmarco_passage_eval msmarco-passage-dev-subset {output_file}'
+        status = os.system(cmd1)
+        stdout, stderr = run_command(cmd2)
+        score = parse_score(stdout, "MRR @10")
+        self.assertEqual(status, 0)
+        self.assertAlmostEqual(score, 0.3584, places=4)
+
+    def test_msmarco_passage_tct_colbert_v2_hnp_bf_bm25_hybrid_otf(self):
+        output_file = 'test_run.msmarco-passage.tct_colbert-v2-hnp.bf-otf.bm25.tsv'
+        self.temp_files.append(output_file)
+        cmd1 = f'python -m pyserini.hsearch dense  --index msmarco-passage-tct_colbert-v2-hnp-bf \
+                                    --encoder castorini/tct_colbert-v2-hnp-msmarco \
+                             sparse --index msmarco-passage \
+                             fusion --alpha 0.12 \
+                             run    --topics msmarco-passage-dev-subset \
+                                    --output {output_file} \
+                                    --batch-size {self.batch_size} --threads {self.threads} \
+                                    --output-format msmarco'
+        cmd2 = f'python -m pyserini.eval.msmarco_passage_eval msmarco-passage-dev-subset {output_file}'
+        status = os.system(cmd1)
+        stdout, stderr = run_command(cmd2)
+        score = parse_score(stdout, "MRR @10")
+        self.assertEqual(status, 0)
+        self.assertAlmostEqual(score, 0.3683, places=4)
+
+    def test_msmarco_passage_tct_colbert_v2_hnp_bf_d2q_hybrid_otf(self):
+        output_file = 'test_run.msmarco-passage.tct_colbert-v2-hnp.bf-otf.doc2queryT5.tsv'
+        self.temp_files.append(output_file)
+        cmd1 = f'python -m pyserini.hsearch dense  --index msmarco-passage-tct_colbert-v2-hnp-bf \
+                                    --encoder castorini/tct_colbert-v2-hnp-msmarco \
+                             sparse --index msmarco-passage-expanded \
+                             fusion --alpha 0.22 \
+                             run    --topics msmarco-passage-dev-subset \
+                                    --output {output_file} \
+                                    --batch-size {self.batch_size} --threads {self.threads} \
+                                    --output-format msmarco'
+        cmd2 = f'python -m pyserini.eval.msmarco_passage_eval msmarco-passage-dev-subset {output_file}'
+        status = os.system(cmd1)
+        stdout, stderr = run_command(cmd2)
+        score = parse_score(stdout, "MRR @10")
+        self.assertEqual(status, 0)
+        self.assertAlmostEqual(score, 0.3730, places=4)
+
+    def test_msmarco_passage_tct_colbert_v2_encoded_queries(self):
+        encoder = QueryEncoder.load_encoded_queries('tct_colbert-v2-msmarco-passage-dev-subset')
+        topics = get_topics('msmarco-passage-dev-subset')
+        for t in topics:
+            self.assertTrue(topics[t]['title'] in encoder.embedding)
+
+    def test_msmarco_passage_tct_colbert_v2_hn_encoded_queries(self):
+        encoder = QueryEncoder.load_encoded_queries('tct_colbert-v2-hn-msmarco-passage-dev-subset')
+        topics = get_topics('msmarco-passage-dev-subset')
+        for t in topics:
+            self.assertTrue(topics[t]['title'] in encoder.embedding)
+
+    def test_msmarco_passage_tct_colbert_v2_hnp_encoded_queries(self):
+        encoder = QueryEncoder.load_encoded_queries('tct_colbert-v2-hnp-msmarco-passage-dev-subset')
+        topics = get_topics('msmarco-passage-dev-subset')
+        for t in topics:
+            self.assertTrue(topics[t]['title'] in encoder.embedding)
+
+    def tearDown(self):
+        clean_files(self.temp_files)
+
+
+if __name__ == '__main__':
+    unittest.main()