Port Machine Learning Module to MLReef
Budget: €30 – €250 EUR
MLReef is the collaboration platform for Machine Learning (ML). Think of it as "GitHub for ML".
It is an open source project that aims at increasing accessibility and adoption of ML.
How?
By introducing AI Modules that are easily discoverable and reusable for ML pipelines - for data preprocessing, data visualization or model training.
Each AI Module is a normal git repository that can be published into "working code", easy to use drag-and-drop elements where user can address parameters and work with them without the need of knowing the code. At the same time it allows a high level of flexibility, as any code can be published containing any function desired. The publishing itself containerizes and registeres the repository with all dependencies and reads out command-line addressable parameters.
What is the objective?
We want to port as many "lying around" ML scripts from GitHub to MLReef. To make them easly useful to our users while still keeping flexibilty and control.
What is the task?
1. Find a ML function of your liking (can be as small as a data-preprocessing opertation to the newest Deel Learning Algorithm) and publish it successfully in MLReef. Take for example:
GitHub: https://github.com/sebastianbk/finetuned-resnet50-keras
MLReef: https://mlreef.com/campac/resnet50
There is a video tutorial on how to port a model. The fast explanation: you need to write decorators based on parameter arguments set by e.g. argparse or any other command line addressable argument function.
2. Show that it worked (via an experiment or data preprocessing or visualization).
We will support you during the entire process. Depending on the code and level of complexity (and your skills), porting a model and publishing it should not take more than 30min - 1h.
We will support you during the entire process, as this is for us also a good task to understand the entire process better.
It is an open source project that aims at increasing accessibility and adoption of ML.
How?
By introducing AI Modules that are easily discoverable and reusable for ML pipelines - for data preprocessing, data visualization or model training.
Each AI Module is a normal git repository that can be published into "working code", easy to use drag-and-drop elements where user can address parameters and work with them without the need of knowing the code. At the same time it allows a high level of flexibility, as any code can be published containing any function desired. The publishing itself containerizes and registeres the repository with all dependencies and reads out command-line addressable parameters.
What is the objective?
We want to port as many "lying around" ML scripts from GitHub to MLReef. To make them easly useful to our users while still keeping flexibilty and control.
What is the task?
1. Find a ML function of your liking (can be as small as a data-preprocessing opertation to the newest Deel Learning Algorithm) and publish it successfully in MLReef. Take for example:
GitHub: https://github.com/sebastianbk/finetuned-resnet50-keras
MLReef: https://mlreef.com/campac/resnet50
There is a video tutorial on how to port a model. The fast explanation: you need to write decorators based on parameter arguments set by e.g. argparse or any other command line addressable argument function.
2. Show that it worked (via an experiment or data preprocessing or visualization).
We will support you during the entire process. Depending on the code and level of complexity (and your skills), porting a model and publishing it should not take more than 30min - 1h.
We will support you during the entire process, as this is for us also a good task to understand the entire process better.