RuntimeError: Can't call numpy() on Tensor that requires grad. Use tensor.detach().numpy() instead.
this error refers to the below sentenceoutput = Parallel(n_jobs=arg.sim.jobs)(delayed(parfun)(i) for i in tqdm.tqdm(range(arg.sim.n_ensemble), position=0, leave=True)) in paragraphif arg.sim.n_ensemble>1: #train network and predict parameters with multiple processes if arg.sim.jobs>1: def parfun(i): net = deep.learn_ML(ML_signal_noisy, arg.sim.bvalues, arg) return deep.predict_ML(ML_signal_noisy[:arg.sim.num_samples_eval, :], arg.sim.bvalues, net, arg) output = Parallel(n_jobs=arg.sim.jobs)(delayed(parfun)(i) for i in tqdm.tqdm(range(arg.sim.n_ensemble), position=0, leave=True)) for bb in range(arg.sim.n_ensemble): paramsNN[aa,bb] = output[bb] #train network and predict parameters with one process
A few days ago, I trained a neural network(NN) with data, which contains 10000 numbers. The NN runs without error but the result is poor.Then I increase the data to 10000000 numbers, expecting the result will be better, but the NN shows the error in a sentence without numpy explicitly.