Thanks @ravivijay that’s a nice idea. One correction: there aren’t separate ‘torch’ and ‘pytorch’ libs in fastai - just one, for which the module is called ‘torch’, but it’s actually referring to pytorch! Pretty confusing… 
Got it. We use only torch. Because torch was modified to work in python, it’s called pytorch.
Thank you!
Thanks for this library description. Does anyone know how I can get the final val_loss (or accuracy) from the learner object after running the fit function? When saving the weights learn.save(‘name’), I want to include the date and the val_loss in the ‘name’, for ease of finding later. Currently, I am doing this:
learn.fit(1e-2, 8, cycle_len=1)
preds,targs=learn.predict_with_targs()
targs_hot = one_hot(targs, preds.shape[1])
loss = preds.shape[1] * (targs_hot * preds).mean()
loss_str = “{:.03f}”.format(loss)
model_name = learn.models.name
now = datetime.datetime.now()
nowstr = now.strftime("%Y-%m-%d-%H-%M-")
learn.save(f’{nowstr}{model_name}{loss_str}dogbreeds’)
learn.load(f’{nowstr}{model_name}{loss_str}dogbreeds’)
I’d rather not have to do the validation prediction and loss calculation again, if the learn object has the loss data saved already. I searched in the metrics and data functions, but couldn’t find it. I also tried to seach the forums for the answer, but couldnt find it. Sorry if it is answered elsewhere.
Thanks again!
Hi. I am new to fast.ai. Will this library support gpu computations ?? Thanks in advance
Could anyone who can tell me why I can not install fastai package?