Recursive Autoencoder

Towards Lossless Encoding of Sentences

Requirements

• Python 3.7
• Pytorch 1.x
• (Optional, for dataset generation) h5py

Training Example

python train.py --dataset_path=<path>

Dataset not included, but dataset_generator.py can be used to generate a hdf5 dataset file from a text file of tokenized sentences, one per line.

Pretrained Model

Embedding size 2048:

model.load_state_dict(torch.load('rae2048.pt'))