Add SPLADE++ ED/SD ONNX regression documentation pages (#2125)

+ Regressions themselves were added in #2094 - this patch adds the documentation. + Fix download paths for SPLADE++ ED/SD corpora. + Minor tweaks for other docs
castorini · May 30, 2023 · a403a2a · a403a2a
1 parent 47d8fde
commit a403a2a
Show file tree

Hide file tree

Showing 49 changed files with 1,446 additions and 84 deletions.
diff --git a/README.md b/README.md
@@ -78,7 +78,9 @@ See individual pages for details!
 | SPLADEv2                                    |       [✓](docs/regressions-msmarco-passage-distill-splade-max.md)        |                                                                       |                                                                       |
 | SPLADE-distill CoCodenser-medium            | [✓](docs/regressions-msmarco-passage-splade-distil-cocodenser-medium.md) | [✓](docs/regressions-dl19-passage-splade-distil-cocodenser-medium.md) | [✓](docs/regressions-dl20-passage-splade-distil-cocodenser-medium.md) |
 | SPLADE++ CoCondenser-EnsembleDistil         |          [✓](docs/regressions-msmarco-passage-splade-pp-ed.md)           |          [✓](docs/regressions-dl19-passage-splade-pp-ed.md)           |          [✓](docs/regressions-dl20-passage-splade-pp-ed.md)           |
+| SPLADE++ CoCondenser-EnsembleDistil (ONNX)  |        [✓](docs/regressions-msmarco-passage-splade-pp-ed-onnx.md)        |        [✓](docs/regressions-dl19-passage-splade-pp-ed-onnx.md)        |        [✓](docs/regressions-dl20-passage-splade-pp-ed-onnx.md)        |
 | SPLADE++ CoCondenser-SelfDistil             |          [✓](docs/regressions-msmarco-passage-splade-pp-sd.md)           |          [✓](docs/regressions-dl19-passage-splade-pp-sd.md)           |          [✓](docs/regressions-dl20-passage-splade-pp-sd.md)           |
+| SPLADE++ CoCondenser-SelfDistil (ONNX)      |        [✓](docs/regressions-msmarco-passage-splade-pp-sd-onnx.md)        |        [✓](docs/regressions-dl19-passage-splade-pp-sd-onnx.md)        |        [✓](docs/regressions-dl20-passage-splade-pp-sd-onnx.md)        |
 
 ### Available Corpora for Download
 

diff --git a/docs/regressions-dl19-passage-splade-distil-cocodenser-medium.md b/docs/regressions-dl19-passage-splade-distil-cocodenser-medium.md
@@ -17,7 +17,7 @@ From one of our Waterloo servers (e.g., `orca`), the following command will perf
 python src/main/python/run_regression.py --index --verify --search --regression dl19-passage-splade-distil-cocodenser-medium
 ```
 
-We make available a version of the MS MARCO passage corpus that has already been processed with SPLADE-distil CoCodenser Medium, i.e., performed model inference on every document and stored the output sparse vectors.
+We make available a version of the MS MARCO Passage Corpus that has already been processed with SPLADE-distil CoCodenser Medium, i.e., performed model inference on every document and stored the output sparse vectors.
 Thus, no neural inference is involved.
 
 From any machine, the following command will download the corpus and perform the complete regression, end to end:

diff --git a/docs/regressions-dl19-passage-splade-pp-ed-onnx.md b/docs/regressions-dl19-passage-splade-pp-ed-onnx.md
@@ -0,0 +1,141 @@
+# Anserini Regressions: TREC 2019 Deep Learning Track (Passage)
+
+**Model**: SPLADE++ CoCondenser-EnsembleDistil (using ONNX for on-the-fly query encoding)
+
+This page describes regression experiments, integrated into Anserini's regression testing framework, using the [SPLADE++ CoCondenser-EnsembleDistil](https://huggingface.co/naver/splade-cocondenser-ensembledistil) model on the [TREC 2019 Deep Learning Track passage ranking task](https://trec.nist.gov/data/deep2019.html), as described in the following paper:
+
+> Thibault Formal, Carlos Lassance, Benjamin Piwowarski, and Stéphane Clinchant. [From Distillation to Hard Negative Sampling: Making Sparse Neural IR Models More Effective.](https://dl.acm.org/doi/10.1145/3477495.3531857) _Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval_, pages 2353–2359.
+
+In these experiments, we are using ONNX to perform query encoding on the fly.
+
+Note that the NIST relevance judgments provide far more relevant passages 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 passage collection, refer to [this page](experiments-msmarco-passage.md).
+
+The exact configurations for these regressions are stored in [this YAML file](../src/main/resources/regression/dl19-passage-splade-pp-ed-onnx.yaml).
+Note that this page is automatically generated from [this template](../src/main/resources/docgen/templates/dl19-passage-splade-pp-ed-onnx.template) as part of Anserini's regression pipeline, so do not modify this page directly; modify the template instead and then run `bin/build.sh` to rebuild the documentation.
+
+From one of our Waterloo servers (e.g., `orca`), the following command will perform the complete regression, end to end:
+
+```bash
+python src/main/python/run_regression.py --index --verify --search --regression dl19-passage-splade-pp-ed-onnx
+```
+
+We make available a version of the MS MARCO Passage Corpus that has already been encoded with SPLADE++ CoCondenser-EnsembleDistil.
+
+From any machine, the following command will download the corpus and perform the complete regression, end to end:
+
+```bash
+python src/main/python/run_regression.py --download --index --verify --search --regression dl19-passage-splade-pp-ed-onnx
+```
+
+The `run_regression.py` script automates the following steps, but if you want to perform each step manually, simply copy/paste from the commands below and you'll obtain the same regression results.
+
+## Corpus Download
+
+Download the corpus and unpack into `collections/`:
+
+```bash
+wget https://rgw.cs.uwaterloo.ca/pyserini/data/msmarco-passage-splade-pp-ed.tar -P collections/
+tar xvf collections/msmarco-passage-splade-pp-ed.tar -C collections/
+```
+
+To confirm, `msmarco-passage-splade-pp-ed.tar` is 4.2 GB and has MD5 checksum `e489133bdc54ee1e7c62a32aa582bc77`.
+With the corpus downloaded, the following command will perform the remaining steps below:
+
+```bash
+python src/main/python/run_regression.py --index --verify --search --regression dl19-passage-splade-pp-ed-onnx \
+  --corpus-path collections/msmarco-passage-splade-pp-ed
+```
+
+## Indexing
+
+Sample indexing command:
+
+```bash
+target/appassembler/bin/IndexCollection \
+  -collection JsonVectorCollection \
+  -input /path/to/msmarco-passage-splade-pp-ed \
+  -index indexes/lucene-index.msmarco-passage-splade-pp-ed/ \
+  -generator DefaultLuceneDocumentGenerator \
+  -threads 16 -impact -pretokenized -storeDocvectors \
+  >& logs/log.msmarco-passage-splade-pp-ed &
+```
+
+The path `/path/to/msmarco-passage-splade-pp-ed/` should point to the corpus downloaded above.
+
+The important indexing options to note here are `-impact -pretokenized`: the first tells Anserini not to encode BM25 doclengths into Lucene's norms (which is the default) and the second option says not to apply any additional tokenization on the SPLADE-distil CoCodenser Medium tokens.
+Upon completion, we should have an index with 8,841,823 documents.
+
+For additional details, see explanation of [common indexing options](common-indexing-options.md).
+
+## Retrieval
+
+Topics and qrels are stored in [`tools/topics-and-qrels/`](../tools/topics-and-qrels/).
+The regression experiments here evaluate on the 43 topics for which NIST has provided judgments 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:
+
+```bash
+target/appassembler/bin/SearchCollection \
+  -index indexes/lucene-index.msmarco-passage-splade-pp-ed/ \
+  -topics tools/topics-and-qrels/topics.dl19-passage.txt \
+  -topicreader TsvInt \
+  -output runs/run.msmarco-passage-splade-pp-ed.splade-pp-ed.topics.dl19-passage.txt \
+  -impact -pretokenized -encoder SpladePlusPlusEnsembleDistil &
+
+target/appassembler/bin/SearchCollection \
+  -index indexes/lucene-index.msmarco-passage-splade-pp-ed/ \
+  -topics tools/topics-and-qrels/topics.dl19-passage.txt \
+  -topicreader TsvInt \
+  -output runs/run.msmarco-passage-splade-pp-ed.rm3.topics.dl19-passage.txt \
+  -impact -pretokenized -encoder SpladePlusPlusEnsembleDistil -rm3 &
+
+target/appassembler/bin/SearchCollection \
+  -index indexes/lucene-index.msmarco-passage-splade-pp-ed/ \
+  -topics tools/topics-and-qrels/topics.dl19-passage.txt \
+  -topicreader TsvInt \
+  -output runs/run.msmarco-passage-splade-pp-ed.rocchio.topics.dl19-passage.txt \
+  -impact -pretokenized -encoder SpladePlusPlusEnsembleDistil -rocchio &
+```
+
+Evaluation can be performed using `trec_eval`:
+
+```bash
+tools/eval/trec_eval.9.0.4/trec_eval -m map -c -l 2 tools/topics-and-qrels/qrels.dl19-passage.txt runs/run.msmarco-passage-splade-pp-ed.splade-pp-ed.topics.dl19-passage.txt
+tools/eval/trec_eval.9.0.4/trec_eval -m ndcg_cut.10 -c tools/topics-and-qrels/qrels.dl19-passage.txt runs/run.msmarco-passage-splade-pp-ed.splade-pp-ed.topics.dl19-passage.txt
+tools/eval/trec_eval.9.0.4/trec_eval -m recall.100 -c -l 2 tools/topics-and-qrels/qrels.dl19-passage.txt runs/run.msmarco-passage-splade-pp-ed.splade-pp-ed.topics.dl19-passage.txt
+tools/eval/trec_eval.9.0.4/trec_eval -m recall.1000 -c -l 2 tools/topics-and-qrels/qrels.dl19-passage.txt runs/run.msmarco-passage-splade-pp-ed.splade-pp-ed.topics.dl19-passage.txt
+
+tools/eval/trec_eval.9.0.4/trec_eval -m map -c -l 2 tools/topics-and-qrels/qrels.dl19-passage.txt runs/run.msmarco-passage-splade-pp-ed.rm3.topics.dl19-passage.txt
+tools/eval/trec_eval.9.0.4/trec_eval -m ndcg_cut.10 -c tools/topics-and-qrels/qrels.dl19-passage.txt runs/run.msmarco-passage-splade-pp-ed.rm3.topics.dl19-passage.txt
+tools/eval/trec_eval.9.0.4/trec_eval -m recall.100 -c -l 2 tools/topics-and-qrels/qrels.dl19-passage.txt runs/run.msmarco-passage-splade-pp-ed.rm3.topics.dl19-passage.txt
+tools/eval/trec_eval.9.0.4/trec_eval -m recall.1000 -c -l 2 tools/topics-and-qrels/qrels.dl19-passage.txt runs/run.msmarco-passage-splade-pp-ed.rm3.topics.dl19-passage.txt
+
+tools/eval/trec_eval.9.0.4/trec_eval -m map -c -l 2 tools/topics-and-qrels/qrels.dl19-passage.txt runs/run.msmarco-passage-splade-pp-ed.rocchio.topics.dl19-passage.txt
+tools/eval/trec_eval.9.0.4/trec_eval -m ndcg_cut.10 -c tools/topics-and-qrels/qrels.dl19-passage.txt runs/run.msmarco-passage-splade-pp-ed.rocchio.topics.dl19-passage.txt
+tools/eval/trec_eval.9.0.4/trec_eval -m recall.100 -c -l 2 tools/topics-and-qrels/qrels.dl19-passage.txt runs/run.msmarco-passage-splade-pp-ed.rocchio.topics.dl19-passage.txt
+tools/eval/trec_eval.9.0.4/trec_eval -m recall.1000 -c -l 2 tools/topics-and-qrels/qrels.dl19-passage.txt runs/run.msmarco-passage-splade-pp-ed.rocchio.topics.dl19-passage.txt
+```
+
+## Effectiveness
+
+With the above commands, you should be able to reproduce the following results:
+
+| **AP@1000**                                                                                                  | **SPLADE++ CoCondenser-EnsembleDistil**| **+RM3**  | **+Rocchio**|
+|:-------------------------------------------------------------------------------------------------------------|-----------|-----------|-----------|
+| [DL19 (Passage)](https://trec.nist.gov/data/deep2020.html)                                                   | 0.5050    | 0.4995    | 0.5140    |
+| **nDCG@10**                                                                                                  | **SPLADE++ CoCondenser-EnsembleDistil**| **+RM3**  | **+Rocchio**|
+| [DL19 (Passage)](https://trec.nist.gov/data/deep2020.html)                                                   | 0.7308    | 0.6849    | 0.7119    |
+| **R@100**                                                                                                    | **SPLADE++ CoCondenser-EnsembleDistil**| **+RM3**  | **+Rocchio**|
+| [DL19 (Passage)](https://trec.nist.gov/data/deep2020.html)                                                   | 0.6390    | 0.6427    | 0.6394    |
+| **R@1000**                                                                                                   | **SPLADE++ CoCondenser-EnsembleDistil**| **+RM3**  | **+Rocchio**|
+| [DL19 (Passage)](https://trec.nist.gov/data/deep2020.html)                                                   | 0.8728    | 0.8684    | 0.8799    |
+
+Note that retrieval metrics are computed to depth 1000 hits per query (as opposed to 100 hits per query for document ranking).
+Also, for computing nDCG, remember that we keep qrels of _all_ relevance grades, whereas for other metrics (e.g., AP), relevance grade 1 is considered not relevant (i.e., use the `-l 2` option in `trec_eval`).
+The experimental results reported here are directly comparable to the results reported in the [track overview paper](https://arxiv.org/abs/2003.07820).
+
+## Reproduction Log[*](reproducibility.md)
+
+To add to this reproduction log, modify [this template](../src/main/resources/docgen/templates/dl19-passage-splade-pp-ed-onnx.template) and run `bin/build.sh` to rebuild the documentation.
diff --git a/docs/regressions-dl19-passage-splade-pp-ed.md b/docs/regressions-dl19-passage-splade-pp-ed.md
@@ -1,11 +1,13 @@
 # Anserini Regressions: TREC 2019 Deep Learning Track (Passage)
 
-**Model**: SPLADE++ CoCondenser-EnsembleDistil
+**Model**: SPLADE++ CoCondenser-EnsembleDistil (using pre-encoded queries)
 
 This page describes regression experiments, integrated into Anserini's regression testing framework, using the [SPLADE++ CoCondenser-EnsembleDistil](https://huggingface.co/naver/splade-cocondenser-ensembledistil) model on the [TREC 2019 Deep Learning Track passage ranking task](https://trec.nist.gov/data/deep2019.html), as described in the following paper:
 
 > Thibault Formal, Carlos Lassance, Benjamin Piwowarski, and Stéphane Clinchant. [From Distillation to Hard Negative Sampling: Making Sparse Neural IR Models More Effective.](https://dl.acm.org/doi/10.1145/3477495.3531857) _Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval_, pages 2353–2359.
 
+In these experiments, we are using pre-encoded queries (i.e., cached results of query encoding).
+
 Note that the NIST relevance judgments provide far more relevant passages 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 passage collection, refer to [this page](experiments-msmarco-passage.md).
 
@@ -18,7 +20,7 @@ From one of our Waterloo servers (e.g., `orca`), the following command will perf
 python src/main/python/run_regression.py --index --verify --search --regression dl19-passage-splade-pp-ed
 ```
 
-We make available a version of the MS MARCO passage corpus that has already been encoded with SPLADE++ CoCondenser-EnsembleDistil.
+We make available a version of the MS MARCO Passage Corpus that has already been encoded with SPLADE++ CoCondenser-EnsembleDistil.
 
 From any machine, the following command will download the corpus and perform the complete regression, end to end:
 
@@ -33,11 +35,11 @@ The `run_regression.py` script automates the following steps, but if you want to
 Download the corpus and unpack into `collections/`:
 
 ```bash
-wget https://rgw.cs.uwaterloo.ca/JIMMYLIN-bucket0/data/msmarco-passage-splade_distil_cocodenser_medium.tar -P collections/
-tar xvf collections/msmarco-passage-splade_distil_cocodenser_medium.tar -C collections/
+wget https://rgw.cs.uwaterloo.ca/pyserini/data/msmarco-passage-splade-pp-ed.tar -P collections/
+tar xvf collections/msmarco-passage-splade-pp-ed.tar -C collections/
 ```
 
-To confirm, `msmarco-passage-splade_distil_cocodenser_medium.tar` is 4.9 GB and has MD5 checksum `f77239a26d08856e6491a34062893b0c`.
+To confirm, `msmarco-passage-splade-pp-ed.tar` is 4.2 GB and has MD5 checksum `e489133bdc54ee1e7c62a32aa582bc77`.
 With the corpus downloaded, the following command will perform the remaining steps below:
 
 ```bash