V0.12.20
TL;DR Run allowed on unrunnable model, fails to not-run properly :D; data wizard table display still broken (info) - but it also appears when using Data Settings on the unrunnable model, where the effect is worse; change dataset doesn’t work even when the alternative is in fact a copy of the original.
Full
The built model is not runnable but Run is still available at the end of the data wizard - that might be OK, but it doesn’t even know that it won’t run even afterwards…
Note that the data wizard column handling still needs fixing: not only is the an issue on 1st use, when invoked again via data settings (2nd image)
There’s also an extra question here: the Perceptilabs post on LinkedIn said
Need to adjust your model’s data or pre-processing settings after you’ve generated it with the Data Wizard?
Click on the Data Settings button on the Modeling Tool’s tool bar at any time to view and modify them
but on the model whose build failed I got this - and though I can’t ignore inputs (index should not be used, my mistake) I can do other stuff
Whereas for a model I had created with more manual input I only get this, to change the train/val/ltest proportions or randomisation:
Note that the Randomize partition check box is invisible and the box is partially transparent
If I then try to change the dataset to a copy of the dataset originally used, the dialog blanks with the spinning wait circle until it finally says “Couldn’t load new dataset”. If I then click that OK to dismiss the error message I am back to the empty dialog and spinning circle with nothing happening… is it reloading the original dataset? It doesn’t say anything… but eventually it does restore the original, though I notice now the path is truncated. Suggestion: user should never see spinning circles without some indication of what is being done otherwise they can only “wait and see”, which isn’t efficient or friendly.