search

probcomp / bdbcontrib

BayesDB contributions, including plotting, helper methods, and examples
http://probcomp.csail.mit.edu/bayesdb
Apache License 2.0
9 stars 6 forks source link

Feature request: 'Zoom in' large heatmaps #126

Open leocasarsa opened 8 years ago

leocasarsa commented 8 years ago

Create an interface to analyze large heatmaps such as the one from the ma-schools dataset.

schools0.heatmap(schools0.q('''ESTIMATE DEPENDENCE PROBABILITY FROM PAIRWISE COLUMNS OF %g'''))

A feature for plotting a heatmap 'on the neighborhood' of an entry in the heatmap, maybe specified by the pair of variables it corresponds to.

The syntax could be ''' schools0.heatmap(query, zoom_in=('SAT math', 'Hispanic'), radius=10) ''' => plots the heatmap around the point of interest and show another entries around the center.

leocasarsa commented 8 years ago

Another important feature would be to make the notion of cluster (that seaborn.clustermap already have) explicit. Then we could plot the heatmap of a specific cluster at a specific level of the dep prob tree.

leocasarsa commented 8 years ago

The current alternative is to zoom manually with matplotlib.

BelhalK commented 8 years ago

I believe a feature with use of the mouse would be better. The solution you suggest implies manually writing down the name of the variables and this already exists with selectors (if using quickstart) or directly in the BQL query with 'WHERE NAME LIKE.."

leocasarsa commented 8 years ago

What I understood with using selectors or "WHERE NAME LIKE" is that you must pass ALL the variables to the selector. What I'm suggesting is that one passes a center (one pair of variables) and a radius only.

Sure, with a mouse would be the optimal solution, but I think that requires a Javascript plugin, which might not be straightforward to implement.

gregory-marton commented 8 years ago

https://github.com/probcomp/bdbcontrib/issues/28 is related.