search

dotnet / machinelearning-modelbuilder

Simple UI tool to build custom machine learning models.
Creative Commons Attribution 4.0 International
264 stars 56 forks source link

Feature: TSV file generation #893

Open Luk164 opened 4 years ago

Luk164 commented 4 years ago

Is your feature request related to a problem? Please describe. Model-builder plugin can automatically create .tsv file from folder structure, but there is no simple way to get this file.

Describe the solution you'd like I would like a way to access the generated .tsv file, even before any learning has been commenced. I currently just go to the TEMP folder and find the latest .tsv file there, but this is a bothersome workaround.

Describe alternatives you've considered A library function that could be called on a folder would be nice too.

Additional context This is related to my question on marketplace:

https://marketplace.visualstudio.com/items?itemName=MLNET.07&ssr=false#qna

I love the simple approach to loading image data, can we get a tool and also a function in code to allow us to get the .tsv file directly from folder? Currently I have to load the data, then go to %appdata%/Local/Temp and find the generated .tsv file if I want to use it.

JakeRadMSFT commented 4 years ago

@Luk164 thanks for providing this feedback! How do you intend to use the TSV? We're trying to figure out how it might fit into Model Builder's flow.

Luk164 commented 4 years ago

Currently machine learning plugin only supports CPU, so I generated the code on almost empty dataset, snatched the TSV of a full dataset I made prior to that, replaced CPU runtime nuget with windows-gpu one and started learning with full dataset on a GPU without writing any code myself.

Luk164 commented 4 years ago

@JakeRadMSFT Forgot to mention you

Luk164 commented 3 years ago

@briacht @JakeRadMSFT Hello, just wanted to check in and see if there was anything being done regarding this. There are not even labels yet on this issue.

JakeRadMSFT commented 3 years ago

@Luk164 Thanks for checking in! We're closer to making this happen than ever :).

Sorry we were in a massive refactoring to help make things like this easier to implement! I'll get this on our nice-to-have list for the next release.