This is a project for interactive data visualization
It uses a dedicated web app with cards that display python variables.
Check out the notebooks folder for examples
Jump right in: Quick start.
from webvis import Vis
vis = Vis(vis_port=7007)
vis.vars.test = "Hello World!"
# Open the browser on 7007 port
vis.show()Then change the name in card to "test", and your variable will appear!
The values are updated dynamically, a separate thread is created that checks the changes.
values can be matplotlib figures, 1-d and 2-d arrays, and even bokeh is supported!
Usecase: machine learning
- Watch and change learning rate
- Watch Loss and accuracy
- Watch output pictures on prediction
- Display confusion matrix
- Control experiments using simple buttons
Advanced:
- Download data of widget
- Filter data to download and send
Feedback from ODS:
- Graph val accuracy vs train accuracy: solved with custom X axis.
- Fast pictures: Solved using legimens module with dict of pictures
- Native support for confusion matrix: Solved by bokeh/matplotlib.
- Storing format of data: Solved by caching utility in front. Also solvable by hook in legimens send
- Control experiments in browser: Solved by button and lock controls.
- Filter data: For God's sake use grep. But what if need to filter by 'greater than'?
- Criteria for outliers: Solved by returning nothing in serializer.
- Large files: Use a dedicated trio coro for every object.