Skip to content
This repository has been archived by the owner on Dec 3, 2024. It is now read-only.

Latest commit

 

History

History
62 lines (45 loc) · 3.04 KB

README.md

File metadata and controls

62 lines (45 loc) · 3.04 KB
Raw

We are deprecating this repository. Please refer to this repository: https://github.com/mederrata/bayesianquilts/tree/main/bayesianquilts/models/spmf

(S)parsley-encoded (P)oisson (M)atrix (F)actorization

Implemented using Tensorflow-probability.

This method differs from conventional hierarchical Poisson Matrix factorization methods primarily by sparsifying the encoding transformation rather than the decoding transformation. The encoding transformation is what computes a representation conditional on data. The decoding transformation takes the representation and produces predictive probability densities. By sparsifying the encoding, we make each representation coordinate a linear combination of a subset of original data features. Hence, inequalities placed on the representation transform directly and transparently into inequalities over the original features.

Installation

Using pip:

pip install git+https://github.com/mederrata/spmf.git

Examples

You'll find these examples under notebooks/ - for your convenience, here are links to open them in Google Colab. Note that you will need to install the package within Colab using

!pip install git+https://github.com/mederrata/spmf.git

  1. Factorization of random noise Open In Colab

  2. Factorization of synthetic data with underlying linear structure Open In Colab

  3. Factorization of synthetic data with underlying nonlinear structure Open In Colab

Factorizing CSV files

We have included a script that you might find to be useful. It is installed into your PATH when using pip.

Usage:

usage: factorize_csv.py [-h] [-f [CSV_FILE]] [-e [EPOCH]] [-d [DIMENSION]]
                        [-b [BATCH_SIZE]] [-lr [LEARNING_RATE]]
                        [-c [CLIP_VALUE]] [-lt] [-rn]

Train PMF on CSV-formatted count matrix

optional arguments:
  -h, --help            show this help message and exit
  -f [CSV_FILE], --csv-file [CSV_FILE]
                        Enter the CSV file
  -e [EPOCH], --epoch [EPOCH]
                        Enter Epoch value: Default: 300
  -d [DIMENSION], --dimension [DIMENSION]
                        Enter embedding dimension. Default: 2
  -b [BATCH_SIZE], --batch-size [BATCH_SIZE]
                        Enter batch size. Default: 5000
  -lr [LEARNING_RATE], --learning-rate [LEARNING_RATE]
                        Enter float. Default: 0.01
  -c [CLIP_VALUE], --clip-value [CLIP_VALUE]
                        Gradient clip value. Default: 3.0
  -lt, --log-transform  Log-transform?
  -rn, --row-normalize  Row normalize based on counts?