stephantul / reach

Load embeddings and featurize your sentences.

Hi there 👋

I'm Stéphan Tulkens! I'm a computational linguistics/AI person. I am currently working as a machine learning engineer at Ecosia.

I got my Phd at CLiPS at the University of Antwerpen under the watchful eyes of Walter Daelemans (Computational Linguistics) and Dominiek Sandra (Psycholinguistics). The topic of my Phd was the way people process orthography during reading. You can find a copy here. Before that I studied computational linguistics (Ma), philosophy (Ba) and software engineering (Ba)

My goal is always to make things as fast and small as possible. I like it when simple models work well, and I love it when simple models get close in accuracy to big models. I do not believe absolute accuracy is a metric to be chased, and I think we should always be mindful of what a model computes or learns from the data.

I’m currently working on 🏃‍♂️:

• reach: a library for loading and working with word embeddings.
• piecelearn: a library that trains a subword tokenizer and embeddings on the same corpus, giving you open vocabulary embeddings.
• unitoken: a library for easy pre-tokenization.
• hashing_split: a library for hash-based data splits (stable splits!)

Other stuff I made (most of it from my Phd) 🐕:

• wordkit: a library for working with orthography
• old20: calculate the orthographic levenshtein distance 20 metric.
• metameric: fast interactive activation networks in numpy.
• humumls: load the UMLS database into a mongodb instance. Fast!
• dutchembeddings: word embeddings for dutch (back when this was a cool thing to do)

My research interests 🤖:

• Tokenizers, specifically subword tokenizers.
• Embeddings, specifically static embeddings (so old-fashioned! 💀), and how to combine these in meaningful ways.
• String similarity, and how to compute it without using dynamic programming.

Contact:

• My website
• Google Scholar
• Twitter

reach's Issues

Add badges

Badges are cool!

Missing dependency: TQDM

This basic usage yields an error:

from reach import Reach

r = Reach.load("path/to/vector")

I think this package needs to add tqdm as dependency, or truly make it optional by importing it optionally.

Add readthedocs, and write the docs to be read

Adding documentation is good

Add Github actions

Adding github actions sounds like a good action

Fix bug: adding <UNK> resets dtype

Loading a .vec file with a given dtype (e.g., float32) using an OOV UNK token silently resets the dtype to the system default (usually float64).

Small fix

Dependabot couldn't authenticate with https://pypi.python.org/simple/

Dependabot couldn't authenticate with https://pypi.python.org/simple/.

You can provide authentication details in your Dependabot dashboard by clicking into the account menu (in the top right) and selecting 'Config variables'.

View the update logs.

tests: rewrite all tests to pytest

All tests are now written in unittest.

OOV

What is meant with "[The] approaches know how to deal with OOV words" ?
Is the deal to delete them?

feat: remove autoreach

BYE 👋

Don't calculate `norm_vectors` if all vectors are already unit length

Even though it is an edge case, it would save 50% memory

Fix string passing in `vector_similarity`

It is still possible to pass strings into the vector_similarity function.

Bump version

Nothing deprecated, so let's bump a minor version

Release 4.1.1

New release to update stuff on pypi

Order of indices

I could look it up myself, but since I am lazy:
If I load the vectors passing wordlist, is it guaranteed that r.vectors entries will be in the same order?
Basically, what I need is the embedding matrix and a vector of words (strings) in correspondence with the words in the matrix (without having to sort again).

Also, the package in pip seems to be outdated :-p

Spin out IO and typing to separate files

Currently everything is one big file. This is dumb!

feat: add support for legacy fast format

Switch to pyproject.toml

We're still using setup, which is dumb. Switch to pyproject.toml

Move from string to hashable

We currently type everything as if all items are strings, but actually they can be any hashable. The typing and docs should be updated to reflect this.

Add pooling functions with safeguard

Pooling is too difficult right now.

problem loading in ipython windows 10 and Ubuntu 20.04

Hi stephan, thanks for making this package!

I've been having trouble importing it on windows 10 and Ubuntu 20.04 in jupyter notebooks: from reach import Reach gives an import error (import reach works, but using rech.Reach.load() doesn't work): unknown location yet the package is visible in the site packages on both machines, under the correct name:

• I git cloned this package into my Anaconda site-packages on both windows 10 and ubuntu
• I checked the python versions in ipython and powershell, they're the same in both windows 10 and ubuntu
• I also checked the PATH variable in both, also the same result in both windows 10 and ubuntu
• I restarted the PCs but it made no difference

I don't think this issue is related to the package versions on my machines (which probably also don't match the package requirements) because the python interpreter doesn't know where to find Reach.

Do you know of any way I could fix this locally?
Thanks anyways!

Best,
Lisa

edit: I can't import it via the python interpreter in the command line either.

docs: rebuild docs + add action that rebuilds docs

Switch to ruff

Switch from flake8 to ruff, update some pre-commit hooks, remove setup.cfg where possible

feat: add new save format

The fast save format is a bit wonky

Expand github actions to also test newer python versions

GH actions currently only tests [3.7, 3.8, 3.9]. Since 3.11 is already released, we should also test that.