dan-gru

comparison.ipynb

In this notebook we:

1. Train sentence representations for a classification task,
2. Probe these representations for analysis, and
3. Perform analysis on trained models.

We will use:

1. A Deep Averaging Network (DAN) model, implemented from scratch,
2. PyTorch's implementation of a GRU based model, and
3. Linear probing models, classifiers that help assess what kinds of information are captured in the sentence representations.