Hey there, trying to use the single-shot prediction with fluid.mlpregressor~, but no success. I use the help file, like this:
predictpoint pointypoint 0 and predictpoint 0 pointypoint, but those result in “Invalid buffer” errors.ignore me
Ah okay. But the reference lists that predictpoint method, so maybe it’ll be there in the future? (feature requeeeeeest)
Otherwise maybe it is possible to hack this functionality by predicting a dataset which only has 1 entry? The question then is whether the weights are reset before each predict message or just “keep rolling”?
You definitely can use predict point. Check the autoencoder example to show how it is done (8c) - it is true that using it on preprocessed data is more useful.
btw @balintlaczko as you learn our framework, can you keep notes of where you stall? That is gold for @weefuzzy @groma and I when we develop learning material to see where someone without our interface decision narrative gets puzzled…
Ooh, I think I just confused it with mds. My bad!
Yeah definitely! I am just putting together an example for time series prediction we are discussing in the other thread. I got stuck at the last (prediction) stage, but now I’ll visit 8c and see how to predict single points. I think it would also be nice to have that as a tab in the fluid.mlpregressor~ help patch.