A library that facilitates a broad set of tools for analysing hidden activations of neural models.
Project description
diagnnose ·
This library contains a set of modules that can be used to analyse recurrent neural networks. In particular, it contains functionality to:
- Extracting activations from different types of (language) models
- Running diagnostic classifiers [1] on extracted activations
- Doing intervention studies [2] with language models
- Analysing word embeddings
- Doing dimensionality reduction and plotting state space trajectories of trained models
Quickstart
Our library is not (yet) officially registered with pip. You can use the library by cloning it and do an editable install with pip:
git clone git@github.com:i-machine-think/diagnnose.git $custom_path
pip3 install -e $custom_path
We will shortly update this README with explanations for the different scripts provided in the library.
Requirements
This library runs with Pytorch 1.0. We refer to the PyTorch website to install the right version for your environment. The preferred version of python is >=3.7.
To install additional requirements, run:
pip install -r requirements.txt
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for diagnnose-0.1a0-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | ea62b3b100d669ab5be0fa34c56cf5b66f1fdc0185409285f54f468b595f07c4 |
|
MD5 | 337c17490c0848954311d676f97035c8 |
|
BLAKE2b-256 | 1af133e6ea274ab8d2677fff01cf04338a0e2ee83eecd6c39a30a8bd7ab148db |