By default tantivy-py uses 128MB heapsize. We change the default to 1GB
and we allow the user to customize this
locally this makes `test_fts.py` run 10x faster
Closes#721
fts will return results as a pyarrow table. Pyarrow tables has a
`filter` method but it does not take sql filter strings (only pyarrow
compute expressions). Instead, we do one of two things to support
`tbl.search("keywords").where("foo=5").limit(10).to_arrow()`:
Default path: If duckdb is available then use duckdb to execute the sql
filter string on the pyarrow table.
Backup path: Otherwise, write the pyarrow table to a lance dataset and
then do `to_table(filter=<filter>)`
Neither is ideal.
Default path has two issues:
1. requires installing an extra library (duckdb)
2. duckdb mangles some fields (like fixed size list => list)
Backup path incurs a latency penalty (~20ms on ssd) to write the
resultset to disk.
In the short term, once #676 is addressed, we can write the dataset to
"memory://" instead of disk, this makes the post filter evaluate much
quicker (ETA next week).
In the longer term, we'd like to be able to evaluate the filter string
on the pyarrow Table directly, one possibility being that we use
Substrait to generate pyarrow compute expressions from sql string. Or if
there's enough progress on pyarrow, it could support Substrait
expressions directly (no ETA)
---------
Co-authored-by: Will Jones <willjones127@gmail.com>
Add `to_list` to return query results as list of python dict (so we're
not too pandas-centric). Closes#555
Add `to_pandas` API and add deprecation warning on `to_df`. Closes#545
Co-authored-by: Chang She <chang@lancedb.com>
I only modified those docs pages that are untouched in existing unmerged
PRs, so hopefully there are no merge conflicts!
1. The `tantivy-py` version specified in the docs doesn't work (pip
install fails), but with the latest version of pip and wheel installed
on my Mac, I was able to just `pip install tantivy` and FTS works great
for me. I updated the docs page to include this in
7ca4b757ce but can always modify to
another specific version in case this breaks any tests.
2. The `.add()` method for Python should take in a list of dicts as the
first option (to also align with the JS API), and additionally, users
can pass an existing pandas DataFrame to add to a table. Hope this makes
sense.
3. I've had multiple conversations with users who are unclear that the
terms "exhaustive", "flat" and "KNN" are all the same kind of search, so
I've updated the verbiage of this section to clarify this.
4. Fixed typos and improved clarity in the ANN indexes page.
- Creates testing files `md_testing.py` and `md_testing.js` for testing
python and nodejs code in markdown files in the documentation
This listens for HTML tags as well: `<!--[language] code code
code...-->` will create a set-up file to create some mock tables or to
fulfill some assumptions in the documentation.
- Creates a github action workflow that triggers every push/pr to
`docs/**`
- Modifies documentation so tests run (mostly indentation, some small
syntax errors and some missing imports)
A list of excluded files that we need to take a closer look at later on:
```javascript
const excludedFiles = [
"../src/fts.md",
"../src/embedding.md",
"../src/examples/serverless_lancedb_with_s3_and_lambda.md",
"../src/examples/serverless_qa_bot_with_modal_and_langchain.md",
"../src/examples/youtube_transcript_bot_with_nodejs.md",
];
```
Many of them can't be done because we need the OpenAI API key :(.
`fts.md` has some issues with the library, I believe this is still
experimental?
Closes#170
---------
Co-authored-by: Will Jones <willjones127@gmail.com>
pypi does not allow packages to be uploaded that has a direct reference
for now we'll just ask the user to install tantivy separately
---------
Co-authored-by: Chang She <chang@lancedb.com>
This is v1 of integrating full text search index into LanceDB.
# API
The query API is roughly the same as before, except if the input is text
instead of a vector we assume that its fts search.
## Example
If `table` is a LanceDB LanceTable, then:
Build index: `table.create_fts_index("text")`
Query: `df = table.search("puppy").limit(10).select(["text"]).to_df()`
# Implementation
Here we use the tantivy-py package to build the index. We then use the
row id's as the full-text-search index's doc id then we just do a Take
operation to fetch the rows.
# Limitations
1. don't support incremental row appends yet. New data won't show up in
search
2. local filesystem only
3. requires building tantivy explicitly
---------
Co-authored-by: Chang She <chang@lancedb.com>
