I’m a beginner to DL and I’ve been trying to use fastai to do a multi-value regression. I’ve 68 values as input and 284 values to output. I’ve been trying to decouple some of the outputs (since there are 4 different probably unrelated groups of values on the output, the first 74, then the next 74, then the next 68 and the last 68) to see if I can get better results.
I’m basically doing this:
splits = RandomSplitter(valid_pct=0.2)(range_of(df)) to = TabularPandas(df, procs=[Normalize], cont_names = ss, y_names=bmbo, splits=splits) dls = to.dataloaders(bs=100) # Create deep learning model learn = tabular_learner(dls, layers=[200,100], metrics=rmse) learn.lr_find() learn.fit_one_cycle(10)
But the validation loss is extremely high (1e9) and goes down slowly in each epoch, so the loss continues high even with 300+ epochs.