Regressions for TREC 2019 Deep Learning Track (Document Ranking Task) (…

…#996) * Regressions for TREC 2019 Deep Learning Track (Document Ranking Task) * Shortened label * More minor tweaks. * CR comments.
castorini · Feb 25, 2020 · 0d42d30 · 0d42d30
1 parent 8350cdd
commit 0d42d30
Show file tree

Hide file tree

Showing 6 changed files with 16,680 additions and 0 deletions.
diff --git a/README.md b/README.md
@@ -74,6 +74,7 @@ Note that these regressions capture the "out of the box" experience, based on [_
 + [Regressions for the MS MARCO Passage Retrieval Task with doc2query expansion](docs/regressions-msmarco-passage-doc2query.md)
 + [Regressions for the MS MARCO Passage Retrieval Task with docTTTTTquery expansion](docs/regressions-msmarco-passage-docTTTTTquery.md)
 + [Regressions for the MS MARCO Document Retrieval](docs/regressions-msmarco-doc.md)
++ [Regressions for the TREC 2019 Deep Learning Track (Document Ranking Task)](docs/regressions-dl19-doc.md)
 + [Regressions for NTCIR-8 ACLIA (IR4QA subtask, Monolingual Chinese)](docs/regressions-ntcir8-zh.md)
 + [Regressions for CLEF 2006 Monolingual French](docs/regressions-clef06-fr.md)
 + [Regressions for TREC 2002 Monolingual Arabic](docs/regressions-trec02-ar.md)

diff --git a/docs/regressions-dl19-doc.md b/docs/regressions-dl19-doc.md
@@ -0,0 +1,115 @@
+# Anserini: Regressions for [DL19 (Document)](https://github.com/microsoft/TREC-2019-Deep-Learning)
+
+This page documents experiments, integrated into Anserini's regression testing framework, for the TREC 2019 Deep Learning Track (Document Ranking Task) on the MS MARCO document collection using relevance judgments from NIST.
+Note that the NIST relevance judgments provide far more relevant documents per topic, unlike the "sparse" judgments provided by Microsoft (these are sometimes called "dense" judgments to emphasize this contrast).
+For additional instructions on working with MS MARCO document collection, refer to [this page](experiments-msmarco-doc.md).
+
+The exact configurations for these regressions are stored in [this YAML file](../src/main/resources/regression/dl19-doc.yaml).
+Note that this page is automatically generated from [this template](../src/main/resources/docgen/templates/dl19-doc.template) as part of Anserini's regression pipeline, so do not modify this page directly; modify the template instead.
+
+## Indexing
+
+Typical indexing command:
+
+```
+nohup sh target/appassembler/bin/IndexCollection -collection TrecCollection -input /path/to/dl19-doc \
+ -index lucene-index.msmarco-doc.pos+docvectors+rawdocs -generator LuceneDocumentGenerator -threads 1 \
+ -storePositions -storeDocvectors -storeRawDocs >& log.dl19-doc.pos+docvectors+rawdocs &
+```
+
+The directory `/path/to/msmarco-doc/` should be a directory containing the official document collection (a single file), in TREC format.
+
+For additional details, see explanation of [common indexing options](common-indexing-options.md).
+
+## Retrieval
+
+Topics and qrels are stored in [`src/main/resources/topics-and-qrels/`](../src/main/resources/topics-and-qrels/).
+The regression experiments here evaluate on the 43 topics for which NIST has provided judgmen as part of the TREC 2019 Deep Learning Track.
+The original data can be found [here](https://trec.nist.gov/data/deep2019.html).
+
+After indexing has completed, you should be able to perform retrieval as follows:
+
+```
+nohup target/appassembler/bin/SearchCollection -index lucene-index.dl19-doc.pos+docvectors+rawdocs \
+ -topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.dl19-doc.txt \
+ -bm25 -output run.dl19-doc.bm25-default.topics.dl19-doc.txt &
+
+nohup target/appassembler/bin/SearchCollection -index lucene-index.dl19-doc.pos+docvectors+rawdocs \
+ -topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.dl19-doc.txt \
+ -bm25 -rm3 -output run.dl19-doc.bm25-default+rm3.topics.dl19-doc.txt &
+
+nohup target/appassembler/bin/SearchCollection -index lucene-index.dl19-doc.pos+docvectors+rawdocs \
+ -topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.dl19-doc.txt \
+ -bm25 -axiom -axiom.deterministic -rerankCutoff 20 -output run.dl19-doc.bm25-default+ax.topics.dl19-doc.txt &
+
+nohup target/appassembler/bin/SearchCollection -index lucene-index.dl19-doc.pos+docvectors+rawdocs \
+ -topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.dl19-doc.txt \
+ -bm25 -bm25prf -output run.dl19-doc.bm25-default+prf.topics.dl19-doc.txt &
+
+nohup target/appassembler/bin/SearchCollection -index lucene-index.dl19-doc.pos+docvectors+rawdocs \
+ -topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.dl19-doc.txt \
+ -bm25 -bm25.k1 3.44 -bm25.b 0.87 -output run.dl19-doc.bm25-tuned.topics.dl19-doc.txt &
+
+nohup target/appassembler/bin/SearchCollection -index lucene-index.dl19-doc.pos+docvectors+rawdocs \
+ -topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.dl19-doc.txt \
+ -bm25 -bm25.k1 3.44 -bm25.b 0.87 -rm3 -output run.dl19-doc.bm25-tuned+rm3.topics.dl19-doc.txt &
+
+nohup target/appassembler/bin/SearchCollection -index lucene-index.dl19-doc.pos+docvectors+rawdocs \
+ -topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.dl19-doc.txt \
+ -bm25 -bm25.k1 3.44 -bm25.b 0.87 -axiom -axiom.deterministic -rerankCutoff 20 -output run.dl19-doc.bm25-tuned+ax.topics.dl19-doc.txt &
+
+nohup target/appassembler/bin/SearchCollection -index lucene-index.dl19-doc.pos+docvectors+rawdocs \
+ -topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.dl19-doc.txt \
+ -bm25 -bm25.k1 3.44 -bm25.b 0.87 -bm25prf -output run.dl19-doc.bm25-tuned+prf.topics.dl19-doc.txt &
+```
+
+Evaluation can be performed using `trec_eval`:
+
+```
+eval/trec_eval.9.0.4/trec_eval -m map -c -m ndcg_cut.10 -c -m recip_rank -c -m recall.100 -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.dl19-doc.txt run.dl19-doc.bm25-default.topics.dl19-doc.txt
+
+eval/trec_eval.9.0.4/trec_eval -m map -c -m ndcg_cut.10 -c -m recip_rank -c -m recall.100 -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.dl19-doc.txt run.dl19-doc.bm25-default+rm3.topics.dl19-doc.txt
+
+eval/trec_eval.9.0.4/trec_eval -m map -c -m ndcg_cut.10 -c -m recip_rank -c -m recall.100 -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.dl19-doc.txt run.dl19-doc.bm25-default+ax.topics.dl19-doc.txt
+
+eval/trec_eval.9.0.4/trec_eval -m map -c -m ndcg_cut.10 -c -m recip_rank -c -m recall.100 -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.dl19-doc.txt run.dl19-doc.bm25-default+prf.topics.dl19-doc.txt
+
+eval/trec_eval.9.0.4/trec_eval -m map -c -m ndcg_cut.10 -c -m recip_rank -c -m recall.100 -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.dl19-doc.txt run.dl19-doc.bm25-tuned.topics.dl19-doc.txt
+
+eval/trec_eval.9.0.4/trec_eval -m map -c -m ndcg_cut.10 -c -m recip_rank -c -m recall.100 -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.dl19-doc.txt run.dl19-doc.bm25-tuned+rm3.topics.dl19-doc.txt
+
+eval/trec_eval.9.0.4/trec_eval -m map -c -m ndcg_cut.10 -c -m recip_rank -c -m recall.100 -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.dl19-doc.txt run.dl19-doc.bm25-tuned+ax.topics.dl19-doc.txt
+
+eval/trec_eval.9.0.4/trec_eval -m map -c -m ndcg_cut.10 -c -m recip_rank -c -m recall.100 -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.dl19-doc.txt run.dl19-doc.bm25-tuned+prf.topics.dl19-doc.txt
+```
+
+## Effectiveness
+
+With the above commands, you should be able to replicate the following results:
+
+MAP                                     | BM25 (Default)| +RM3      | +Ax       | +PRF      | BM25 (Tuned)| +RM3      | +Ax       | +PRF      |
+:---------------------------------------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|
+[DL19 (Doc)](https://trec.nist.gov/data/deep2019.html)| 0.3309    | 0.3870    | 0.3516    | 0.3624    | 0.3138    | 0.3697    | 0.3860    | 0.3858    |
+
+
+NDCG@10                                 | BM25 (Default)| +RM3      | +Ax       | +PRF      | BM25 (Tuned)| +RM3      | +Ax       | +PRF      |
+:---------------------------------------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|
+[DL19 (Doc)](https://trec.nist.gov/data/deep2019.html)| 0.5190    | 0.5169    | 0.4730    | 0.5105    | 0.5140    | 0.5485    | 0.5245    | 0.5280    |
+
+
+RR                                      | BM25 (Default)| +RM3      | +Ax       | +PRF      | BM25 (Tuned)| +RM3      | +Ax       | +PRF      |
+:---------------------------------------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|
+[DL19 (Doc)](https://trec.nist.gov/data/deep2019.html)| 0.8046    | 0.7718    | 0.7428    | 0.7775    | 0.8872    | 0.8074    | 0.7492    | 0.8007    |
+
+
+R@100                                   | BM25 (Default)| +RM3      | +Ax       | +PRF      | BM25 (Tuned)| +RM3      | +Ax       | +PRF      |
+:---------------------------------------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|
+[DL19 (Doc)](https://trec.nist.gov/data/deep2019.html)| 0.3948    | 0.4189    | 0.3945    | 0.4004    | 0.3862    | 0.4193    | 0.4399    | 0.4287    |
+
+
+R@1000                                  | BM25 (Default)| +RM3      | +Ax       | +PRF      | BM25 (Tuned)| +RM3      | +Ax       | +PRF      |
+:---------------------------------------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|
+[DL19 (Doc)](https://trec.nist.gov/data/deep2019.html)| 0.6966    | 0.7504    | 0.7323    | 0.7357    | 0.6810    | 0.7282    | 0.7545    | 0.7553    |
+
+The setting "default" refers the default BM25 settings of `k1=0.9`, `b=0.4`, while "tuned" refers to the tuned setting of `k1=3.44`, `b=0.87` (see [this page](experiments-msmarco-doc.md) for more details about tuning).
+
diff --git a/src/main/resources/docgen/templates/dl19-doc.template b/src/main/resources/docgen/templates/dl19-doc.template
@@ -0,0 +1,47 @@
+# Anserini: Regressions for [DL19 (Document)](https://github.com/microsoft/TREC-2019-Deep-Learning)
+
+This page documents experiments, integrated into Anserini's regression testing framework, for the TREC 2019 Deep Learning Track (Document Ranking Task) on the MS MARCO document collection using relevance judgments from NIST.
+Note that the NIST relevance judgments provide far more relevant documents per topic, unlike the "sparse" judgments provided by Microsoft (these are sometimes called "dense" judgments to emphasize this contrast).
+For additional instructions on working with MS MARCO document collection, refer to [this page](experiments-msmarco-doc.md).
+
+The exact configurations for these regressions are stored in [this YAML file](../src/main/resources/regression/dl19-doc.yaml).
+Note that this page is automatically generated from [this template](../src/main/resources/docgen/templates/dl19-doc.template) as part of Anserini's regression pipeline, so do not modify this page directly; modify the template instead.
+
+## Indexing
+
+Typical indexing command:
+
+```
+${index_cmds}
+```
+
+The directory `/path/to/msmarco-doc/` should be a directory containing the official document collection (a single file), in TREC format.
+
+For additional details, see explanation of [common indexing options](common-indexing-options.md).
+
+## Retrieval
+
+Topics and qrels are stored in [`src/main/resources/topics-and-qrels/`](../src/main/resources/topics-and-qrels/).
+The regression experiments here evaluate on the 43 topics for which NIST has provided judgmen as part of the TREC 2019 Deep Learning Track.
+The original data can be found [here](https://trec.nist.gov/data/deep2019.html).
+
+After indexing has completed, you should be able to perform retrieval as follows:
+
+```
+${ranking_cmds}
+```
+
+Evaluation can be performed using `trec_eval`:
+
+```
+${eval_cmds}
+```
+
+## Effectiveness
+
+With the above commands, you should be able to replicate the following results:
+
+${effectiveness}
+
+The setting "default" refers the default BM25 settings of `k1=0.9`, `b=0.4`, while "tuned" refers to the tuned setting of `k1=3.44`, `b=0.87` (see [this page](experiments-msmarco-doc.md) for more details about tuning).
+
diff --git a/src/main/resources/regression/dl19-doc.yaml b/src/main/resources/regression/dl19-doc.yaml
@@ -0,0 +1,216 @@
+---
+name: msmarco-doc
+index_command: target/appassembler/bin/IndexCollection
+index_utils_command: target/appassembler/bin/IndexUtils
+search_command: target/appassembler/bin/SearchCollection
+topic_root: src/main/resources/topics-and-qrels/
+qrels_root: src/main/resources/topics-and-qrels/
+index_root:
+ranking_root:
+collection: TrecCollection
+generator: LuceneDocumentGenerator
+threads: 1
+index_options:
+  - -storePositions
+  - -storeDocvectors
+  - -storeRawDocs
+topic_reader: TsvInt
+evals:
+  - command: eval/trec_eval.9.0.4/trec_eval
+    params:
+      - -m map
+      - -c
+    separator: "\t"
+    parse_index: 2
+    metric: map
+    metric_precision: 4
+    can_combine: true
+  - command: eval/trec_eval.9.0.4/trec_eval
+    params:
+      - -m ndcg_cut.10
+      - -c
+    separator: "\t"
+    parse_index: 2
+    metric: NDCG@10
+    metric_precision: 4
+    can_combine: true
+  - command: eval/trec_eval.9.0.4/trec_eval
+    params:
+      - -m recip_rank
+      - -c
+    separator: "\t"
+    parse_index: 2
+    metric: RR
+    metric_precision: 4
+    can_combine: true
+  - command: eval/trec_eval.9.0.4/trec_eval
+    params:
+      - -m recall.100
+      - -c
+    separator: "\t"
+    parse_index: 2
+    metric: R@100
+    metric_precision: 4
+    can_combine: true
+  - command: eval/trec_eval.9.0.4/trec_eval
+    params:
+      - -m recall.1000
+      - -c
+    separator: "\t"
+    parse_index: 2
+    metric: R@1000
+    metric_precision: 4
+    can_combine: true
+input_roots:
+  - /tuna1/      # on tuna
+  - /store/      # on orca
+  - /scratch2/   # on damiano
+input: collections/msmarco/doc/
+index_path: indexes/lucene-index.msmarco-doc.pos+docvectors+rawdocs
+index_stats:
+  documents: 3213835
+  documents (non-empty): 3213835
+  total terms: 2748636047
+topics:
+  - name: "[DL19 (Doc)](https://trec.nist.gov/data/deep2019.html)"
+    path: topics.dl19-doc.txt
+    qrel: qrels.dl19-doc.txt
+models:
+  - name: bm25-default
+    display: BM25 (Default)
+    params:
+      - -bm25
+    results:
+      map:
+        - 0.3309
+      NDCG@10:
+        - 0.5190
+      RR:
+        - 0.8046
+      R@100:
+        - 0.3948
+      R@1000:
+        - 0.6966
+  - name: bm25-default+rm3
+    display: +RM3
+    params:
+      - -bm25
+      - -rm3
+    results:
+      map:
+        - 0.3870
+      NDCG@10:
+        - 0.5169
+      RR:
+        - 0.7718
+      R@100:
+        - 0.4189
+      R@1000:
+        - 0.7504
+  - name: bm25-default+ax
+    display: +Ax
+    params:
+      - -bm25
+      - -axiom
+      - -axiom.deterministic
+      - -rerankCutoff 20
+    results:
+      map:
+        - 0.3516
+      NDCG@10:
+        - 0.4730
+      RR:
+        - 0.7428
+      R@100:
+        - 0.3945
+      R@1000:
+        - 0.7323
+  - name: bm25-default+prf
+    display: +PRF
+    params:
+      - -bm25
+      - -bm25prf
+    results:
+      map:
+        - 0.3624
+      NDCG@10:
+        - 0.5105
+      RR:
+        - 0.7775
+      R@100:
+        - 0.4004
+      R@1000:
+        - 0.7357
+  - name: bm25-tuned
+    display: BM25 (Tuned)
+    params:
+      - -bm25
+      - -bm25.k1 3.44
+      - -bm25.b 0.87
+    results:
+      map:
+        - 0.3138
+      NDCG@10:
+        - 0.5140
+      RR:
+        - 0.8872
+      R@100:
+        - 0.3862
+      R@1000:
+        - 0.6810
+  - name: bm25-tuned+rm3
+    display: +RM3
+    params:
+      - -bm25
+      - -bm25.k1 3.44
+      - -bm25.b 0.87
+      - -rm3
+    results:
+      map:
+        - 0.3697
+      NDCG@10:
+        - 0.5485
+      RR:
+        - 0.8074
+      R@100:
+        - 0.4193
+      R@1000:
+        - 0.7282
+  - name: bm25-tuned+ax
+    display: +Ax
+    params:
+      - -bm25
+      - -bm25.k1 3.44
+      - -bm25.b 0.87
+      - -axiom
+      - -axiom.deterministic
+      - -rerankCutoff 20
+    results:
+      map:
+        - 0.3860
+      NDCG@10:
+        - 0.5245
+      RR:
+        - 0.7492
+      R@100:
+        - 0.4399
+      R@1000:
+        - 0.7545
+  - name: bm25-tuned+prf
+    display: +PRF
+    params:
+      - -bm25
+      - -bm25.k1 3.44
+      - -bm25.b 0.87
+      - -bm25prf
+    results:
+      map:
+        - 0.3858
+      NDCG@10:
+        - 0.5280
+      RR:
+        - 0.8007
+      R@100:
+        - 0.4287
+      R@1000:
+        - 0.7553