TL;DR Me Gusta Mucho!
- Yes, the data wizard now handles more inputs correctly - that’s a great help
- For my redshift estimator - the (regression) model built was immediately viable (though loss could have defaulted to quadratic rather than cross-entropy) and the merge component nicely concatenated
- Still loving the new UI theme and OS native file navigation - it’s silky smooth now
- I have not found any major issues.
Really nice guys! Thanks
A Feature Request
One thing that would really help with the training graphs is provision of a “zoom” capability (see this Feature Request - please upvote!). At the beginning of training the loss is high and whatever is happening after epoch 1 is imperceptible - this is a grab of the current graph
That said, the ability to enlarge individual graphs from this sort of view is very nice
What a difference being able to enlarge/zoom makes to seeing the detail of e.g. R Squared
- If “Stop” button is clicked in Training view, the model seems to stop, but on returning to the model view, the button at top right next to “data settings” still says “Stop” rather than Run (from the stopped state) (which to me means that the Model button isn’t properly linked to the state)
- If the “Pause” button is clicked in Training view it continues to show the pause symbol where I would expect the play symbol (right pointing triangle)
- Although the data wizard now handles larger number of inputs, if there are many inputs not all buttons at the top of the training view are accessible without scaling the browser: suggestion - scrollable list to the right of the loss graphs (i.e. make them shorter; length also matters less when there is zoom)
- When “data” is graphed, it would be helpful if there were a legend for the one-hot encoding - in the example below there are 13 named “surveys” but I have no idea which one was assigned “8”
I’m sure clicking on images in the forum used to pop them up; that doesn’t happen any more, which is a pity.