Crate sif_embedding

sif-embedding

This crate provides simple but powerful sentence embedding techniques based on Smooth Inverse Frequency (SIF) described in the paper:

Sanjeev Arora, Yingyu Liang, and Tengyu Ma, A Simple but Tough-to-Beat Baseline for Sentence Embeddings, ICLR 2017.

Specifications

This crate depends on ndarray-linalg. You must always specify which backend will be used with features, following the specifications of ndarray-linalg. See README of ndarray-linalg v0.16.0 since the feature names of sif-embedding are the same.

For example, you can specify the OpenBLAS backend as follows:

# Cargo.toml

[dependencies]
sif-embedding = { version = "0.1", features = ["openblas"] }

If you are having problems compiling this library due to the backend, my tips may help.

Basic usage

See the document of Sif.

Re-exports

• pub use lexicon::Lexicon;
• pub use sif::FreezedSif;
• pub use sif::Sif;
• pub use word_embeddings::WordEmbeddings;

Modules

• lexicon
  Handlers for vocabulary.
• sif
  Smooth Inverse Frequency (SIF).
• util
  Utilities.
• word_embeddings
  Handlers for word embeddings from pretrained word2vec/GloVe models.

Type Definitions

• Float
  Common type of floating numbers.