xiazh18/learngit

fsklearn

A simple module for machine learning in Fortran using scikit-learn.

Overview

• Currently the module can be used to do training and prediction in FORTRAN.

• The training part uses scikit-learn library by calling Python from FORTRAN.

• The prediction part, as it may be called frequent by a FORTRAN code, is written with FORTRAN 2003.

Supported machine learning method

Currently, the following methods are supported for regression (not classification) problem:

• Neual Networks
• Decision Tree
• Random Forest

Supported FORTRAN compiler

The following compiler are tested.

• Intel - Tested with 2019.0.2.187
• GNU - Tested with 8.1.0

Since derived type is used in the module, it is recommended to used GNU > 5.0 or Intel > 14.0 (2013 SP1).

Required packages for Python3

• numpy
• json
• scikit-learn

build tool

• Makefile
• FoBis.py (testing)

Run the test

    git clone https://github.com/Yeldon/fsklearn.git
    cd fsklearn/
    cp -r tests/src ./
    cp -r tests/build ./
    make 
    ./build/fsklearn_test

Example

Assume you have set up the file path and the correct input and output interface for your data, a simple main program (sequential version) could be:

• training

    program main
    use mod_fsklearn
    implicit none
    integer :: input_len = 3
    integer :: input_len = 3
    integer :: num_datalen = 500
    integer :: sample_data(500,6)
    
    Call fsklearn_initialization ! Initialization
    ! Assume you have defined your interface in mod_fsklearn
    Call F_Sklearn%Fen_Training(sample_data,3, 3, 500)
    Call F_Sklearn%PY_Training ! Call python to train
    
    end program main

• prediction

    program main
    use mod_fsklearn
    implicit none
    integer :: inputs = 3
    
    Call fsklearn_initialization ! Initialization
    inputs=[-0.99,0.141067, -0.54]
    ! If you have setup the mod_fsklearn and input.namelist, this should be working
    F_Sklearn%outputs = F_Sklearn%predict(inputs,F_Sklearn%n_inputs,F_Sklearn%n_outputs)

Progress[6/8]

• Glue FORTRAN and Python
• Basic training and prediction interface
• MPI version
• First step tests
• A more smart Python code generator
• Parameters for training function
• Prediction for data, vector and matrix
• Second step tests

License

This project is licensed under the BSD3 - see the LICENSE.md file for details