Features

• Ease of Use
• Speed
• Expandability
• Simplicity

• Ease of Use

  PHOTON offers an intuitive approach to designing complex machine learning pipelines from scratch in under 3 minutes. Be productive right from the start using e.g. our preloaded sklearn pipelines or Deep Learning model architectures.

• Speed

  While modern machine learning enables mind-blowing applications, a vast, ever increasing hyperparameter space needs to be searched to optimize machine learning pipelines. PHOTON meets this challenge in two ways:

  PHOTON enables intelligent, nested hyperparameter search minimizing the number of function evaluations necessary.

  It also allows you to define your own performance expectations in order to rapidly decide if a given configuration is promising or not, not wasting time testing bad configurations.

• Expandability

  Machine Learning is moving at a neck-breaking pace today. As exciting new algorithms and architectures appear almost every week, we explicitly built PHOTON in a way that allows you to easily integrate with the software you are already using.

  PHOTON is designed for cross plattform support, allowing you to take full advantage of all your favorite machine learning packages simultaneously.

  Add your favorite elements from any other software package such as preprocessing steps, metrics, and machines using PHOTON’s simple, standardized register.

  Also hook up your own neural network models, preprocessing steps or even completely new hyperparameter optimizers using our sklearn-based wrapper interface.

  Everything you add is immediately accessible in all your models.

• Simplicity

  While terrific machine learning tools exist, they focus on the models themselves, mostly neglecting the workflow necessary for rapid prototyping. PHOTON was consequently designed for rapid prototyping, allowing you to build, train and evaluate complete machine learning analyses fast.