Open agitter opened 5 years ago
Microsoft has drag and drop machine learning interfaces, described here:
Turi Create doesn't seem to have a graphical interface, but it was recommended to me by a machine learning novice. This could be a potential resource for someone who can write Python code and wants more wrappers and assistance than scikit-learn provides.
DataChat: https://datachat.ai/
A good summer exercise may be to spend a little time with each of these tools to see whether we would recommend them to participants once they have learned basic ML concepts.
anaconda just released Voilà - according to the one line summary
Voilà turns Jupyter notebooks into standalone web applications.
Here are the links https://blog.jupyter.org/and-voil%C3%A0-f6a2c08a4a93 and https://anaconda.org/conda-forge/voila
I looked at Voilà briefly, and my first impression is that this is different from Binder because it strips away all of the code and displays only the output of code cells. That output can include interactive plots and widgets, creating a simple interactive interface for users who do not want to see the code. The example also looks like it is running locally (as opposed to Binder running the notebook remotely), but I didn't check carefully.
Voilà could be appealing for our demo notebook. It would be an intermediate layer between the ml4bio GUI and the full notebook with scikit-learn code.
If Voilà runs locally, we would still need users to configure a conda environment on their machine. One appealing aspect of Binder and similar services is they do not have any local dependencies.
Yes, it looks like it runs locally.
Makes sense! I will play around with some of these this week.
@cmilica found https://www.streamlit.io/ It builds interactive machine learning apps with compact Python scripts. The gallery has lots of cool examples.
Milica has also been looking into interactive visualizations that teach machine learning concepts, such as https://thomas-tanay.github.io/post--L2-regularization/
Galaxy-ML is a relevant resource:
Gradio generates GUIs for machine learning models
PennAI has a graphical interface and supports AutoML:
The interface supports saving a trained model and exporting a script to reproduce the results, which we've discussed.
Thought this was really cool https://climateprimer.mit.edu/
@cmilica found http://www.r2d3.us/ with great visualizations for decision trees and bias-variance tradeoffs.