macavaney/d2q-msmarco-passage-scores-electra

---
annotations_creators:
- no-annotation
language: []
language_creators:
- machine-generated
license: []
pretty_name: Doc2Query ELECTRA Relevance Scores for `msmarco-passage`
source_datasets: [msmarco-passage]
tags:
- document-expansion
- doc2query--
task_categories:
- text-retrieval
task_ids:
- document-retrieval
viewer: false
---

# Doc2Query ELECTRA Relevance Scores for `msmarco-passage`

This dataset provides the pre-computed query relevance scores for the [`msmarco-passage`](https://ir-datasets.com/msmarco-passage) dataset,
for use with Doc2Query--.

The generated queries come from [`macavaney/d2q-msmarco-passage`](https://huggingface.co/datasets/macavaney/d2q-msmarco-passage) and
were scored with [`crystina-z/monoELECTRA_LCE_nneg31`](https://huggingface.co/crystina-z/monoELECTRA_LCE_nneg31).

## Getting started

This artefact is meant to be used with the [`pyterrier_doc2query`](https://github.com/terrierteam/pyterrier_doc2query) pacakge. It can
be installed as:

```bash
pip install git+https://github.com/terrierteam/pyterrier_doc2query
```

Depending on what you are using this aretefact for, you may also need the following additional packages:

```bash
pip install git+https://github.com/terrierteam/pyterrier_pisa # for indexing / retrieval
pip install git+https://github.com/terrierteam/pyterrier_dr # for reproducing this aretefact
```

## Using this artefact

The main use case is to use this aretefact in a Doc2Query−− indexing pipeline:

```python
import pyterrier as pt ; pt.init()
from pyterrier_pisa import PisaIndex
from pyterrier_doc2query import QueryScoreStore, QueryFilter

store = QueryScoreStore.from_repo('https://huggingface.co/datasets/macavaney/d2q-msmarco-passage-scores-electra')
index = PisaIndex('path/to/index')
pipeline = store.query_scorer(limit_k=40) >> QueryFilter(t=store.percentile(70)) >> index

dataset = pt.get_dataset('irds:msmarco-passage')
pipeline.index(dataset.get_corpus_iter())
```

You can also use the store directly as a dataset to look up or iterate over the data:

```python
store.lookup('100')
# {'querygen': ..., 'querygen_store': ...}
for record in store:
pass
```

## Reproducing this aretefact

This aretefact can be reproduced using the following pipeline:

```python
import pyterrier as pt ; pt.init()
from pyterrier_dr import ElectraScorer
from pyterrier_doc2query import Doc2QueryStore, QueryScoreStore, QueryScorer

doc2query_generator = Doc2QueryStore.from_repo('https://huggingface.co/datasets/macavaney/d2q-msmarco-passage').generator()
store = QueryScoreStore('path/to/store')
pipeline = doc2query_generator >> QueryScorer(ElectraScorer()) >> store

dataset = pt.get_dataset('irds:msmarco-passage')
pipeline.index(dataset.get_corpus_iter())
```

Note that this process will take quite some time; it computes the relevance score for 80 generated queries
for every document in the dataset.

原始信息汇总

数据集概述

数据集名称

• 名称: Doc2Query ELECTRA Relevance Scores for msmarco-passage

数据集来源

• 源数据集: msmarco-passage

数据集用途

• 用途: 提供预计算的查询相关性分数，用于Doc2Query--。

数据集内容

• 生成查询来源: macavaney/d2q-msmarco-passage
• 评分模型: crystina-z/monoELECTRA_LCE_nneg31

数据集使用

• 主要使用场景: 在Doc2Query--索引管道中使用。
• 示例代码: 使用pyterrier_doc2query包进行数据集的处理和索引。

数据集复现

• 复现流程: 需要使用pyterrier_dr包中的ElectraScorer进行相关性分数的计算。
• 计算量: 为数据集中的每个文档计算80个生成查询的相关性分数。