Some days ago, I discovered a very interesting cloud-based platform, named Neuton, which permits the building of Machine Learning models, without any coding or Machine Learning knowledge. In practice, you can develop a high-performance model just in three simple steps:
- Upload a dataset
- Train the model *Make predictions
In this article, I give an overview of Neuton, how it works and an example of its usage. A free version of Neuton is available here, thus you can try it :)
1 Overview of Neuton
Neuton is a cloud-based platform, which allows the development of high-performance Machine Learning models without any knowledge of Machine Learning. Compared to other existing solutions, Neuton does not require any software or library installation. The only requirement is to create an account in the platform and start to use it.
According to the authors, Neuton is based on a Neural Network (NN) unique patented algorithm which automatically grows the Neural Network from a single neuron during training with constant cross-validation and stops just before the model is overfit. So there is no predefined NN architecture! Exciting, right?
In addition, models deployed in Neuton can be easily embedded on tiny devices because of their compact size.
The interesting aspect of Neuton is its ability to organize itself, without the external intervention of the data scientist, for data preprocessing, parameter tuning and so on. Neuton even warns if a model has become obsolete and needs to be trained on a fresh set of data.
2 Working with Neuton
When you log in to Neuton, you see a very user-friendly interface. You can add a New Solution (i.e. a project) at the beginning of your journey. Once you’ve created a new solution, you need to perform just three simple steps: dataset, training, prediction.
The three steps in Neuton: dataset, training, prediction
2.1 Dataset
The first step involves dataset uploading. Your dataset must meet Dataset Requirements, which include (but not limited to) the following aspects:
- the dataset must be a CSV file, where the first row must indicate the column names;
- the file format must be utf-8.
It does not matter that the dataset has been already preprocessed with data preprocessing or feature engineering techniques. It is sufficient to upload a reasonably organized dataset, which may also contain missing values, duplicates, etc. Anyway, Neuton also provides an Advanced Mode, where you can perform basic manipulation, enable/disable data preprocessing and feature engineering, training timing limitations and Time Series analysis settings.
The next step involves Target Variable selection, among the available columns of the dataset. Then you can choose the task type (Regression, Binary Classification or Multiclass Classification), as well as the metric you want to consider.
2.2 Training
Training in Neuton is the simplest task. All you need to do is to press the button Start Training.
An ad-hoc virtual machine is created in the cloud, to perform your training task. Training may take some time. However, there is a progress bar which indicates the percentage of completion. It is also possible to receive a message on the phone when the training operation is finished.
2.3 Prediction
Once a model has been trained, the corresponding virtual machine is shutdown and the model can be either used to perform predictions or downloaded to be deployed for production.
Continue reading on Towards Data Science
Top comments (0)