biolab
Big-idea issues (a collection)
This is not a graveyard of issues.
To prevent stalled issues with big ideas (that can be implemented only as somebody's summer-time pet project) from becoming the white noise of the issue list, such issues are closed, but referenced here, so they are collected and can be opened any time.
Quite doable, with some dedication
**Input data viewer in the tree widget (#7082): Tree viewer could accept data and show how it is classified (distribution by nodes) or a classification path for a single data instance.
Better saving of graphs (#4306): Allow user to specify additional properties prior to saving, for instance axis titles.
ODBC Support (#3565): Doable, but requires a willing developer. The issue hasn't stirred much interest in the core team, so outside help very welcome.
Heatmap: Creation of groups (#2983) and Showing selected clusters (#2902): Probably requires refactoring of heatmap widget.
Parameter tuning (e.g. grid search) for SVM, trees ... (#1327): This is a long-term wish, but we don't know how to implement it generally enough.
OpenML Data set widget (#1265): Nothing complicated, but in more than three years nobody sat down and implemented it.
Improvements in Heatmap (#4380): Annotations etc.
Density scatter plot (#4884).
Bayes net structural learning (#4972).
Scatter plot matrices (#5229)
Scatter Plot and/or t-SNE: Show more than 10 clusters (#3679): Requires nice graphical design and a lot of love. Looks nice on a picture densely covered with points, but probably won't on common scatter plots.
3d Scatter Plot (#5546): This is a persistent wish, but we question its utility. The implementation must show that it can do something worth doing, which Linear Projection widge can't do.
Forward selection (#6480) in widgets like Linear Projection and Radviz.
Doable futuristic projects
Export Schema to Python Code (#3978): the referred PR contains an idea that could be implemented. This project requires a lot of careful designing, and discussion of seasoned developers.
Histograms on top of columns / Data wrangler (#3882): Fancy, with lots of user interaction. Complicated to implement.
MDS via computeValue (#3802): I suppose this requires a reimplementation of MDS with fixed embedding. Not not doable, but needs a strong champion.
HDF5 (#945): Basic readers are trivial, while full support requires a well-designed widget or widgets that can accept structured data.
"Mega widgets" (#4910): These can take different forms, either as widget that contains a canvas and defines output, or it can be just within GUI - the user would be able to collapse widget into a black hole and expand it back.
Regarding: 3d Scatter Plot is a persistent wish, but we question its utility.
There are entire domains of data that are inherently 3D which cannot reasonably be displayed as 2D projections. Paths of flying animals or objects, anything which is a function of 2 input variables. An easy out might be to create a plotly widget.
For anyone looking for 3D Scatter Plot, I offer the following: https://github.com/iraytrace/orange3_scatter3d