Anserini CLI

Use Anserini CLI

Overview

Use this skill when Anserini is already available through either a resolved fatjar or a source checkout. This skill covers command usage, not environment setup or builds. If no usable fatjar or checkout is present, use $install-anserini-fatjar or $install-anserini-dev-env first.

Do not run commands that trigger large prebuilt-index downloads unless the user explicitly asks for retrieval experiments or index downloads.

Examples below use the fatjar form:

java -cp "$ANSERINI_JAR" <main-class> <args>

From an Anserini source checkout, replace java -cp "$ANSERINI_JAR" with bin/run.sh:

bin/run.sh <main-class> <args>

Keep commands pinned to the same jar or checkout unless the user asks to change versions.

Runtime Check

For a fatjar workflow, confirm ANSERINI_JAR is set and points to an existing jar:

test -n "$ANSERINI_JAR"
test -f "$ANSERINI_JAR"

For a checkout workflow, confirm bin/run.sh is available:

test -x bin/run.sh

A useful functional smoke test is:

java -cp "$ANSERINI_JAR" io.anserini.search.SearchCollection \
  -threads 1 \
  -index cacm \
  -topics cacm \
  -output run.cacm.bm25.txt \
  -hits 1000 \
  -bm25

This command may download the small CACM prebuilt index and topics on first use.

Prebuilt Index Registry

To inspect prebuilt indexes exposed by io.anserini.cli.PrebuiltIndexRegistry, run:

java -cp "$ANSERINI_JAR" io.anserini.cli.PrebuiltIndexRegistry --list

--list emits JSON in current jars, so prefer --filter and jq instead of grepping raw output. If jq is not available, ask the user whether it should be installed before relying on jq examples.

msmarco-v1-passage is a common choice and should be called out when users ask about available prebuilt indexes or MS MARCO passage retrieval setup.

Recommended lookup for the standard MS MARCO V1 passage inverted index:

java -cp "$ANSERINI_JAR" io.anserini.cli.PrebuiltIndexRegistry --list --filter '^msmarco-v1-passage$' \
  | jq '.[0] | {name, type, description, filename}'

Useful variants:

java -cp "$ANSERINI_JAR" io.anserini.cli.PrebuiltIndexRegistry --help
java -cp "$ANSERINI_JAR" io.anserini.cli.PrebuiltIndexRegistry --list --filter 'msmarco.*passage' | jq '.[].name'
java -cp "$ANSERINI_JAR" io.anserini.cli.PrebuiltIndexRegistry --type flat --list
java -cp "$ANSERINI_JAR" io.anserini.cli.PrebuiltIndexRegistry --type inverted --list
java -cp "$ANSERINI_JAR" io.anserini.cli.PrebuiltIndexRegistry --type impact --list
java -cp "$ANSERINI_JAR" io.anserini.cli.PrebuiltIndexRegistry --type hnsw --list

Topics Registry

To inspect topics exposed by io.anserini.cli.TopicsRegistry, run:

java -cp "$ANSERINI_JAR" io.anserini.cli.TopicsRegistry --list

--list emits canonical topic names as JSON. Use --filter <regexp> to narrow the registry with a regular expression:

java -cp "$ANSERINI_JAR" io.anserini.cli.TopicsRegistry --list --filter 'msmarco' | jq '.'

--get writes parsed topics as JSON to stdout:

java -cp "$ANSERINI_JAR" io.anserini.cli.TopicsRegistry --get <name>

--metadata writes the canonical name, aliases, registered path, reader class, and downloaded local path as JSON to stdout:

java -cp "$ANSERINI_JAR" io.anserini.cli.TopicsRegistry --metadata <name>

Both --get and --metadata download the registered topics when needed.

For the standard MS MARCO V1 passage queries that pair with the msmarco-v1-passage prebuilt index, use the canonical name msmarco-passage.dev-subset:

java -cp "$ANSERINI_JAR" io.anserini.cli.TopicsRegistry --metadata msmarco-passage.dev-subset | jq '.'

Aliases such as msmarco-v1-passage.dev are also accepted by --get and --metadata.

Qrels Registry

To inspect qrels exposed by io.anserini.cli.QrelsRegistry, run:

java -cp "$ANSERINI_JAR" io.anserini.cli.QrelsRegistry --list

--list emits canonical qrels names as JSON. Use --filter <regexp> to narrow the registry with a regular expression:

java -cp "$ANSERINI_JAR" io.anserini.cli.QrelsRegistry --list --filter 'msmarco' | jq '.'

--get writes raw qrels to stdout:

java -cp "$ANSERINI_JAR" io.anserini.cli.QrelsRegistry --get <name>

--metadata writes the canonical name, aliases, registered path, and downloaded local path as JSON to stdout:

java -cp "$ANSERINI_JAR" io.anserini.cli.QrelsRegistry --metadata <name>

Both --get and --metadata download the registered qrels when needed.

For the standard MS MARCO V1 passage relevance judgments that pair with the msmarco-v1-passage prebuilt index, use the canonical name msmarco-passage.dev-subset:

java -cp "$ANSERINI_JAR" io.anserini.cli.QrelsRegistry --metadata msmarco-passage.dev-subset | jq '.'

Search CLI

Use io.anserini.cli.Search for ad hoc retrieval against either a local Lucene index path or a prebuilt index name.

Example using the popular msmarco-v1-passage prebuilt index:

java -cp "$ANSERINI_JAR" io.anserini.cli.Search --index msmarco-v1-passage --query "what is a lobster roll" --hits 10 --json

Interactive mode:

java -cp "$ANSERINI_JAR" io.anserini.cli.Search --index msmarco-v1-passage --interactive --json

Useful output variants:

java -cp "$ANSERINI_JAR" io.anserini.cli.Search --index msmarco-v1-passage --query "what is a lobster roll" --json
java -cp "$ANSERINI_JAR" io.anserini.cli.Search --index msmarco-v1-passage --query "what is a lobster roll" --trec

Get Document CLI

Use io.anserini.cli.GetDocument to fetch the stored raw document for a collection docid from either a local Lucene index path or a prebuilt index name.

Example using the popular msmarco-v1-passage prebuilt index:

java -cp "$ANSERINI_JAR" io.anserini.cli.GetDocument --index msmarco-v1-passage --docid 2161721

Interactive mode reads docids from stdin:

java -cp "$ANSERINI_JAR" io.anserini.cli.GetDocument --index msmarco-v1-passage --interactive

This command prints the document's stored raw field. It reports an error when the docid is not found or when the index does not store raw documents.

SearchCollection

Use io.anserini.search.SearchCollection for batch retrieval over a topic set. It writes TREC run files and supports retrieval-model flags such as -bm25, -rm3, -rocchio, -hits, and -threads. Use io.anserini.cli.Search instead for single-query or interactive inspection.

Canonical CACM example using a prebuilt index and built-in topic symbol:

java -cp "$ANSERINI_JAR" io.anserini.search.SearchCollection \
  -index cacm \
  -topics cacm \
  -output run.cacm.bm25.txt \
  -hits 1000 \
  -bm25

Evaluate the CACM run with Anserini's Java trec_eval wrapper:

java -cp "$ANSERINI_JAR" io.anserini.eval.TrecEval \
  -c \
  -m map \
  -m P.30 \
  cacm \
  run.cacm.bm25.txt

Expected scores are MAP 0.3123 and P30 0.1942.

To verify them mechanically:

java -cp "$ANSERINI_JAR" io.anserini.eval.TrecEval \
  -c \
  -m map \
  -m P.30 \
  cacm \
  run.cacm.bm25.txt | tee eval.cacm.bm25.txt

grep -Eq '^map[[:space:]]+all[[:space:]]+0\.3123$' eval.cacm.bm25.txt
grep -Eq '^P_30[[:space:]]+all[[:space:]]+0\.1942$' eval.cacm.bm25.txt

REST API Server

Use io.anserini.api.RestServer to expose search and document lookup over HTTP.

Fatjar invocation:

java -cp "$ANSERINI_JAR" io.anserini.api.RestServer --port 8081

Sample requests against the popular msmarco-v1-passage index:

curl "http://localhost:8081/v1/msmarco-v1-passage/search?query=what%20is%20anserini&hits=5"
curl "http://localhost:8081/v1/msmarco-v1-passage/doc/2161721"

This REST workflow is most useful when users want to query the same prebuilt indexes exposed by the CLI, especially msmarco-v1-passage.

Troubleshooting

• No fatjar found: use $install-anserini-fatjar to download a released Maven Central fatjar, or $install-anserini-dev-env if the user needs a jar built from the source checkout.
• Missing bin/run.sh: use $install-anserini-dev-env from an Anserini checkout.
• ClassNotFoundException: confirm the jar or checkout was built from the expected Anserini version.
• RestServer reports Port already in use for unused ports in a sandboxed Codex session: local socket binding may be blocked by sandbox permissions. Rerun the server command with escalation, and use an available high local port if the documented port is occupied.