A better machine learning "model package" than Keras: no preprocessing required, 0 code to build models

Xiaoxiao sent from Aofei Temple
Quantum Bit Report | Public Account QbitAI

When making machine learning models, have you used up all your brain cells just by integrating various algorithms?

Or do you think that data preprocessing is a "waste of time"?

A guy who graduated from the University of Göttingen and works on machine learning also discovered this problem: he originally just wanted to design a model, but it turned out that "implementation was more troublesome than designing."

So he created a project called igel (which means "hedgehog" in German, but is also the abbreviation of Init, Generate, and Evaluate Machine Learning) , which allows him to design the desired model without writing any extra code.

In other words, users only need to know the principles of various machine learning models, but do not need to write code by themselves.

Compared to Keras, this project further simplifies the work of preprocessing, input and output, making it as simple as a "model package".

The author spent two weeks to produce version 1.0 of the project, which received 842 stars within one day and is still rising rapidly.

Since it is only version 1.0, the author said that this project still has a lot of room for improvement.

But for now, it is enough for basic machine learning modeling.

“Automating machine learning production”

The original intention of the author to establish this project was to provide machine learning models to everyone.

Whether you are a technical engineer or a non-technical user in other industries, you can use machine models to make your work easier.

To put it simply, it is to turn machine learning into an "automated" process, and he designed it this way.

The igel project contains all the latest machine learning models (regression, classification, clustering) , and the author said that the project is still being updated.

In other words, if there are newer machine learning models in the future, they will be added to the project.

The model currently supports the following functions:

Supports all the latest machine learning models (even preview models)

Support different data preprocessing methods

Provides flexibility and data control when writing configuration

Support cross validation

Support yaml and json formats

Support for different sklearn metrics for regression, classification and clustering

Supports multi-output/multi-target regression and classification

Supports multi-processing parallel model building

It can be seen that the author has worked hard to make machine learning production more concise .

If you want to get started, it's very easy.

6 steps to get started with the "Model Pack"

Like other programs, the author provides a "Help" menu for this program. You only need to enter "igel -h" (or igel -help) to learn how to use it.

After learning how to use it, you can start creating configuration files, whether in yaml or json format.

If you are a lazy person (like the author) , you can use "igel init" to initialize.

For example, if you want to make a configuration file for the function of judging whether you are sick:

The first step is to select the function, model, and target: igel init -type "classification" -model "NeuralNetwork" -target "sick"

The second step is initialization: igel init

Then, the program will generate a configuration file for you and modify it as needed.

After that, it’s a matter of choosing the specific algorithm parameters and providing a dataset of your choice.

For example, if you want to use random forest to process data, you only need to provide parameters to the system (as well as the dataset and configuration file path) , and it will help you train:

In addition, you can evaluate the model/pre-trained model:

Feeling good? Generate a prediction model:

Use it directly:

In just 6 simple steps, the machine model (preview model) has been generated, which is very convenient.

Netizen: Great, I want more new features

In addition, the author is also very happy to adopt the opinions of netizens.

For example, one user pointed out that it would be a great improvement if cross-validation could be used for hyperparameter search/tuning, and the author immediately adopted this suggestion.

However, some netizens also said that it is meaningless to "automate all procedures."

After all, the author has stated that the project is built on the basis of scikit-learn. But the latter can also do machine learning with a few lines of code, so what is the difference between scikit-learn and this project?

The author responded that the biggest difference is that compared to writing code, this project aims to design the desired model in a more "readable" way.

Some netizens agree with this point of view. After all, for many machine learning engineers working in production, "any programming work that does not need to be done" is meaningless and is a "waste of time and money."

More netizens expressed their support for this project and hoped to see its new features.

Friends who are interested in this machine learning project can check it out through the portal below~

about the author

Nidhal Baccouri obtained his master's degree from the University of Göttingen in Germany in April this year, with research interests in software, control engineering and artificial intelligence.

Currently, Nidhal Baccouri works in the automotive industry, researching digital twin technology, applying knowledge of artificial intelligence and the Internet of Things. In his spare time, he likes to work on projects, especially those related to AI and Python.

Portal

Project address:
https://github.com/nidhaloff/igel

Reference links:
https://news.ycombinator.com/item?id=24671525
https://nidhalbacc.azurewebsites.net/