SHALO

Shallow baseline models for text in Tensorflow

Implemented models

Baseline models

• SparseLM: Linear model trained over sparse bag-of-words representation
• LSTM: Long short-term memory model
• LSTMPreTrain: LSTM with pretrained embeddings

Mean word vector models

• LinearModel: Linear model over fixed, pretrained embeddings
• fastText: Implementation of fastText in Tensorflow
• fastTextPreTrain: fastText initialized with pretrained embeddings

"Tough-to-Beat Baseline" models

• TTBB: Implementation of A Simple but Tough-to-Beat Baseline for Sentence Embeddings in Tensorflow
• TTBBTune: TTBB with gradient-tuned embeddings, common component, and smoothing parameter
• TTBBTuneExact: TTBB with gradient-tuned embeddings and smoothing parameter, with common component updated exactly

Getting started

Install dependencies

pip install --requirement python-package-requirements.txt

Get data

scripts/get_senna.sh
scripts/get_word_freq.sh

Run tests for simple models

python shalo_test.py

Notes

Test data is derived from Pang and Lee (2005).