msurtsukov / neural-ode Goto Github PK

Something I haven't been able to get my head around is the need to find d(a)/dt. Isn't the gradient of the loss function enough for backpropagation? Sorry if it's a trivial doubt.

Getting the following error.

RuntimeError Traceback (most recent call last)
in ()
----> 1 conduct_experiment(ode_true, ode_trained, 500, "linear")

in conduct_experiment(ode_true, ode_trained, n_steps, name, plot_freq)
38
39 optimizer.zero_grad()
---> 40 loss.backward(retain_graph=True)
41 optimizer.step()
42

/usr/local/lib/python3.6/dist-packages/torch/tensor.py in backward(self, gradient, retain_graph, create_graph)
91 products. Defaults to False.
92 """
---> 93 torch.autograd.backward(self, gradient, retain_graph, create_graph)
94
95 def register_hook(self, hook):

/usr/local/lib/python3.6/dist-packages/torch/autograd/init.py in backward(tensors, grad_tensors, retain_graph, create_graph, grad_variables)
88 Variable._execution_engine.run_backward(
89 tensors, grad_tensors, retain_graph, create_graph,
---> 90 allow_unreachable=True) # allow_unreachable flag
91
92

RuntimeError: Function ODEAdjointBackward returned an invalid gradient at index 2 - expected shape [4] but got [1, 4]

Hello, I want to ask about the last element of s0 in the picture of algorithm 1. Why the initial value of at(t1) is -dL/dt1 but not dL/dt1? I don't understand the minus sign here.

Hello, I think the Eq. 2 and Eq.4 is not correct.