totvslabs/pyCarol

Create bokeh dashboard for neural networks based models

Closed this issue · 4 comments

rafonseca commented

We started thi idea in carol-ds-education, but we thought that would be nice to have a common bokeh dashboard to analyze our models. The idea here is to use it mainly for intenal analysis, but I think many front-end ideas mayl come from this dashboard.

rafonseca commented

The initial proposed layout is the following (https://github.com/totvslabs/carol-ds-education/issues/67):

dashboard should show:
(a) t-SNE plot of all current students, colored according to score
(b) side selectors:

  • course
  • period
  • score

(c) selected students features impact (if possbile with features values)
(d) selectable score histogram of selected students
(e) datatable
layout:

_______
|  a    b   |
|  c    d   |
|  e    e   |
rafonseca commented

Of course, this implies in another task, that is to build a bokeh plots library. bokeh is a low-level API and many of the common plots should be implemented manually, but once the plot functions are done we can reuse them.

rafonseca commented

moving this to pyscorch