Question Running Regression on Barracuda

Hello,

**Update**

We are running a tensorflow regression model in barracuda. We have converted the model into a frozen_graph and then to a ONNX - we have gotten some predictions and have realized that they are not scaled. We are using joblib to pickle our scalers - is there a way to bring them into barracuda?

 

Hi, I am relatively new to Unity and am also attempting to do regression type analysis/prediction with Barracuda. I have been able to follow Classification tutorials that involved converting my tensorflow model to ONNX, then uploading that to Unity and writing a "Classify" method within the C# script. Basically it can recognize a dog and a cat. However, what I am really needing to do is more related to regression. If there are any tutorials or example code you could point me to I would be incredibly grateful for your help. By the way this is purely for educational purposes and I am not making any $$$. Thanks very much for your consideration.

 

Hi fguinier, thanks for the reply. I am not necessarily new to ML, but definitely so in the Unity / Barracuda environment. I have experience programming basic perceptrons and MLPs from scratch and beyond that I have mostly used Tensorflow in Google CoLab and similar higher level tools like WEKA and DeepLearning4j. I will definitely take a look at that course, but just from a quick glance at the material, I am at least familiar with the concepts. (No doubt I could use a refresher!) Is there any more 'Unity specific' type tutorial or course you could recommend? Thanks again for the quick reply!​