ML Notebook improvements

[luisquintanilla]

Thanks @andrasfuchs for trying out the notebooks and providing feedback:

This feedback is specific to the Training and AutoML notebook, but it's also good feedback in general to follow across notebooks.

• [ ] The capitalization of "ML.NET" is not consistent in the document
• [x] The "This notebook is under active development" is barely visible with Visual Studio's dark theme
• [ ] The tutorial is easy to follow, I just missed the links to the different classes and methods (like .Fit, IDataView, Regression.Trainers, TrainTestSplit, etc.), because although they are familiar it would be nice to have a way to jump to that particular class or method to investigate it further. F1 doesn't seem to work as it normally does when I browse my own code.
• [ ] It would have been a little better if the sample datasets had more than one feature columns, just to see how Transforms.Concatenate should look like in that case

• [x] In the "Use AutoML..." section when I run the code, it only showed a grey rectangle while it was running, and when it finished it remained a grey and made all the other previous outputs shown as grey rectangles.

  

  When I closed the file, reopened it and clicked "Run All", I got the following exception:

  

[luisquintanilla]

@LittleLittleCloud do you know why the grey box / exception might be happening?

[JakeRadMSFT]

I believe it's related to throughput issue with Daytona/WebView2. Oddly enough ... I think it mostly happens on very powerful machines.

@andrasfuchs is your machine fairly powerful?

I thought I decreased the occurrence of this issue but I'll need to dig in more.

[JakeRadMSFT]

Also thanks so much for trying it out @andrasfuchs!! We'll work to resolve that feedback in the Notebooks.

[andrasfuchs]

@JakeRadMSFT Relatively yes, 6-core Ryzen 5 3600 CPU, 32 GB RAM, NVIDIA RTX 3070 Ti GPU. I'm not sure it matters, but I have many background processes running normally and my GPU is constantly under 90%+ load.

I don't know if it's related, but I noticed before a few times that during some Model Builder training sessions in Visual Studio the AutoML trainer started to eat up all the memory that I have left, and Windows started to hang more and more, so I needed to terminate the training process. I couldn't figure out how to reliably reproduce it though.

[JakeRadMSFT]

@andrasfuchs can you try out the latest version of the notebooks? Perf should be improved and the grey boxes should be gone :)

[andrasfuchs]

@JakeRadMSFT I'd love to, but I'm abroad for a few weeks and I have limited internet access. I'll get back to you when I'm home again.

[andrasfuchs]

@JakeRadMSFT Today I run the code snippets in the notebook again, and it got significantly better. The UI is more responsive, it feels quicker too.

I noticed four small, low-priority things:

• [x] One of the sample codes run for 300 seconds. I think that's a little too much since it's just a tech demo, so I think ~30 seconds would give a better UX.
• [ ] There is a relatively wide table under the chart, and I think the horizontal scrollbar is a little difficult to find. The values in its last Parameters column are difficult to read, it would be nice to have a pretty formatting for that parameters array.
• [ ] Scrolling with the mouse wheel doesn't work if the cursor is above one of the results (under the code snippets). You can test this by running a few of the snippets, go to the top of the page, and start scrolling down by using the mouse wheel. It will stop working as soon as you reach a section with results.
• [ ] When I scroll down in the notebook without clicking into it, and I switch another window (with alt+tab) and come back again to Visual Studio, the notebook loses it's last scrolling position and I'm back at the top of the notebook. If I clicked into a section and scrolled further down before switching then when I switch back the notebook jumps to the clicked section.

I also experienced a Windows-hanging situation again that I suspect to be a memory-leak within the notebook, but I want to investigate it a little further. I think that is a more serious problem, so it's worth a separate issue. I'll create a new issue for that if I'm able to reproduce it a few more times.