What is your experience with the models from "rtdl"?

[Yura52]

Hi! Thank you for your interest in rtdl! We suggest sharing feedback on the models from the rtdl library in this discussion. This discussion can help users make decisions whether they should try some model(s) in their tasks.

You can cover anything you think is important. For example:

• have you successfully applied some model from rtdl to solve some task?
• what was the task type? What was the dataset size? What was the number of features? What was the domain?
• what models did you compare? Did you compare them with Gradient Boostings? What was the result in terms of target metrics and other aspects? (efficiency, API, etc.)
• if you tried to use some model from rtdl, but then decided to use other solution, what were the reasons?
• and anything else!

[thelexhibition]

Hi, I am interested in trying the algorithm but noticed the example in Colab only has an example for how it is used in a regression task and it is not clear to me how to modify it to work for classification. Is it possible to provide a simple example of it used in classification for the Forest Cover Type dataset for example? Thanks

[Yura52]

We have extended the notebook with a classification example. The only thing you need is to uncomment the corresponding code in the cell where data is prepared. Note that the classification dataset that we added is quite big and you will need GPU/TPU in order to train FT-Transformer in a reasonable time. However, CPU should be enough for MLP/ResNet.