# experimental-mf

**Repository Path**: mirrors_dmlc/experimental-mf

## Basic Information

- **Project Name**: experimental-mf
- **Description**: cache-friendly multithread matrix factorization
- **Primary Language**: Unknown
- **License**: Apache-2.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2020-08-08
- **Last Updated**: 2026-07-18

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Fast Matrix Factorization

## Features
   * Cache-friendly Multithread Matrix Factorization.
   * Fast Multithread Stochastic Gradient Langevin Dynamics (SGLD) for Matrix Factorization.
   * Fast Multithread Differentially Private Matrix Factorization.
   * Matrix Factorization with Adaptive Regularizer.

## Data
We support Google's Protobuf as input. Try data/getdata.cc to convert from userwise raw data to protobuf format:

    ./getdata -r [userwise_raw_data] -w [protobuf_binary] --method [protobuf] --size [int]
  A sample of userwise raw data looks like:

    0:
    11,5.0
    21,3.0
    1:
    9,5.0
    12,1.0

  where there are two users '0' and '1'.

  Or if you only have rating wise raw data, you can first convert to a userwise raw data:

    ./getdata -r [rating_wise_raw] -w [userwise_raw_data] --method [userwise] --split [int]

  A sample of rating_wise_raw data with a header looks like:

    100000
    0,1,5.0
    0,2,1.0
  where the header indicates the number of ratings, and follows by user_id, item_id and rating in each line.

## Environment Requirment
  * GCC 4.9 or higher

    ```bash
    tar zxf gcc-4.9.2.tar.gz;cd gcc-4.9.2;contrib/download_prerequisites;cd ..;mkdir buildc;cd buildc;../gcc-4.9.2/configure --disable-multilib;make -j 32;sudo make install;cd ..;
    ```
  * Intel TBB

    ```bash
    sudo apt-get install libtbb-dev
    ```
  * Google Protobuf

    ```bash
    sudo apt-get install -y libprotobuf-dev; sudo apt-get install -y protobuf-compiler;
    ```
  * Intel MKL


## Reference
[1] [Fast Differentially Private Matrix Factorization](http://arxiv.org/abs/1505.01419). Ziqi Liu, Yu-Xiang Wang, Alex Smola.

[2] Learning recommender systems with adaptive regularization. Steffen Rendle.