FeatureScreeningDemo.jl

Installation

You need Julia >= v1.6.0.

add "ssh://git@github.com/cursorinsight/FeatureScreeningDemo.jl"#master

Set up for non-julia users

Install Julia v >= 1.6.0. You can install it with asdf.

Clone the project

$ git clone git@github.com:cursorinsight/FeatureScreeningDemo.jl.git
$ cd FeatureScreeningDemo.jl/
$ julia --project

julia>                                   -- Press ']'
(FeatureScreeningDemo) pkg> instantiate
(FeatureScreeningDemo) pkg> build
(FeatureScreeningDemo) pkg>              -- Press Backspace

Good to have

Install hdf5-tools to get a lot different tools for HDF5 files. For example:

• h5dump: enables the user to examine the contents of an HDF5 file and dump those contents in human readable form

$ sudo apt install hdf5-tools

Usage

This application demonstrates the usage and usefulness of our feature screening method. You can run a complete demo to test all functionality, run some benchmark to measure your feature set or screen you feature set.

Create a random feature dataset

If you don't have a valid dataset to test the features, you can generate random training feature set.

using FeatureScreening.Types: FeatureSet, save
feature_set = rand(FeatureSet, 25, 200; label_count = 5)
save(feature_set)

CLI usage

You can use our implementation from command line too. To get the proper usage of the commands, see the next section about the --help option.

-h|--help option

You can run julia --project src/main.js command with -h|--help option to get a short description about the application and the proper usage.

demo command

This command demonstrates most of the features of this demo application.

1. Generate random train feature set
2. Screen the train data set
3. Run benchmarks on
   • the original train feature set,
   • the hypothetical best subset of the original feature set and
   • the screened subset.

Usage:

$ julia --project src/main.jl demo

benchmark command

This command benchmarks the given training feature set.

Usage:

$ julia --project src/main.jl \
    benchmark \
    [--config <config>] \
    [--output <result_dictionary>] \
    <training-data-hdf5> \
    [<test-data-hdf5>]

If you don't have valid dataset, you can generate random training feature set. You can find the steps for that under Create random feature dataset heading.

screen command

This command screens the given feature set and generate a screened one.

Usage:

$ julia --project src/main.js \
    screen \
    [--config <config>] \
    [--output <result_dictionary>] \
    <training-data-hdf5>

If you don't have valid dataset, you can generate random training feature set. You can find the steps for that under Create random feature dataset heading.