christophsk / classifier-lit

PAIR Code's Language Interpretability Tool (LIT) for Text Classification

🔥 classifier-lit The Language Interpretability Tool (LIT) for Text Classification

This is an implementation of the PAIR-code Language Interpretability Tool for text classification. This was assembled from examples as a way to experiment with various features of LIT. See the LIT User's Guide for more information.

Quickstart

Install the requirements,

pip install -r requirements

run the (bash) script

./run_demo.sh

The required pytorch model files will be downloaded if they are not cached

starting LIT server, model cardiffnlp/twitter-roberta-base-sentiment
I0925 15:00:12.167857 8647302656 seq_model.py:86] RobertaForSequenceClassification loaded for 3 labels
I0925 15:00:12.174917 8647302656 seq_dataset.py:62]          rows:   20
I0925 15:00:12.176527 8647302656 seq_dataset.py:63] unique labels:    3
I0925 15:00:12.177896 8647302656 dev_server.py:88]
 (    (
 )\ ) )\ )  *   )
(()/((()/(` )  /(
 /(_))/(_))( )(_))
(_)) (_)) (_(_())
| |  |_ _||_   _|
| |__ | |   | |
|____|___|  |_|


I0925 15:00:12.177957 8647302656 dev_server.py:89] Starting LIT server...
I0925 15:00:12.177999 8647302656 caching.py:124] CachingModelWrapper 'classifier': no cache path specified, not loading.
I0925 15:00:12.178436 8647302656 gradient_maps.py:120] Skipping token_grad_sentence since embeddings field not found.
I0925 15:00:12.178502 8647302656 gradient_maps.py:235] Skipping token_grad_sentence since embeddings field not found.
I0925 15:00:12.178730 8647302656 wsgi_serving.py:41]

Starting Server on port 5432
You can navigate to 127.0.0.1:5432


I0925 15:00:12.179692 8647302656 _internal.py:225]  * Running on http://127.0.0.1:5432/ (Press CTRL+C to quit)

Paste 127.0.0.1:5432 into your browser.

Model

A model name or path for a transformers SequenceClassification model. This can be a path to your (pytorch) trained model or the name an appropriate model from HuggingFace models.

Data

The data is a .csv, ideally consisting of validation data, with one column for the text and the other validation label.

If the validation label is not known, use 0 (zero). The metrics will be meaningless, but all the other LIT features are available.

Using a GPU

A GPU is automatically detected and used. If your GPU instance is a remote, headless cloud instance, you can still use your local browser by using port forwarding feature of SSH. For the default port 5432:

ssh -i access-creds.pem -L 5432:localhost:5432 <your id>@<remote IP address>

Start the server on the remote and then view the results in your local browser.

Commandline Usage

usage: seq_server.py [-h] --model_path MODEL_PATH --data_path DATA_PATH
                     --label_text_cols LABEL_TEXT_COLS
                     [--batch_size BATCH_SIZE] [--max_seq_len MAX_SEQ_LEN]
                     [--port PORT]

Start the LIT server

optional arguments:
  -h, --help            show this help message and exit
  --model_path MODEL_PATH
                        tar.gz, name, or directory of the pytorch model
  --data_path DATA_PATH
                        path + file.csv, for the data .csv
  --label_text_cols LABEL_TEXT_COLS
                        python-style list of indexes [label-index, text-index]
                        in the .csv
  --batch_size BATCH_SIZE
                        batch size, default=8
  --max_seq_len MAX_SEQ_LEN
                        maximum sequence length up to 512, default=128
  --port PORT           LIT server port, default=5432

License

MIT License Copyright © 2021 Chris Skiscim

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE

lit-nlp is licensed under the Apache License Version 2.0