AutoML-Benchmark

Benchmark for some usual automated machine learning, such as: AutoSklearn, MLJAR, H2O, TPOT and AutoGluon. All visualized via a responsive Dash Ploty Web Application.

Installation

First of all install the python3-pip package and then the virtual environment for safety reason:

sudo apt install python3-pip
sudo apt install default-jre
sudo pip3 install virtualenv 

Then create a new virtual environment:

virtualenv my_venv

Access at it and activate it:

cd my_venv
source bin/activate.fish

Clone my repository and install all dependencies with:

make install

Usage

To run the app execute the following line of code:

python3 start.py

Open your favourite browser and go to: http://127.0.0.1:8050/. Here you will be albe to interact with the application

Type of tests

There are five types of operations:

1. OpenML Benchmark: Here you have two options:

   1. You can insert a sequrnce of dataframe ID where each ID is followed by a comma. This command will run a benchmark on the inserted sequence
   2. Or you can choose the number of dataframes each for classification task and for regression task and the number of instances that a dataframe at least has. This command will start a benchmark using openml dataframes

2. Kaggle Benchmark: Here you can choose multiple kaggle's dataframes for running a benchmark on them

3. Test Benchmark: Here you can run a benchmark on a specific dataframe by insering the dataframe id and using a single algorithm ot all of them by selecting a options

4. Past Results OpenML: Here you can navigate between past OpenML benchmark by selecting a specific date, or you can comapre multiple OpenML Benchmarks that have the same dataframes but with different timelife

5. Past Results Kaggle: Here you can navigate between past Kaggle benchmark by selecting a specific date, or you can comapre multiple Kaggle Benchmarks that have the same dataframes but with different timelife

Actions available

In all operation these action are available:

• Analize the results of Classification Tasks and Regression Tasks by a Table visualization, Bar Charts visualization and Scatter Plot visualization
• See the Timelife of all algorithms
• Inspect the Pipeline of al algorithms