torchdyn

torchdyn copied to clipboard

Dynamical system on collections of tensors

2

A dynamical system can often be described not by a single tensor but by multiple ones. For example, a system of particles can have node features, edge features, and global...

SimonKitSangChu

Torchscripting not possible with `NeuralODE` due to function redefinition

5

**Describe the bug** Torchscript is unable to script the `NeuralODE` class due to a function being redefined. This is a problem because there is control flow present in the code...

StephenHogg

Clone parts of the `odeint` and ODE solver implementations to get an internal, simplified `sdeint` API. Complete the `EulerMaruyama` placeholder in `solvers.py`.

Zymrael

Standardize the two classes. `NeuralSDE`s should have an option to use our internals or `torchsde` as a backend. Related to [this ](https://github.com/DiffEqML/torchdyn/issues/97) for implementation of our internal solver SDE suite.

Zymrael

I am trying to train a Neural ODE in my local gpu to speed up my training. However, for my surprise it slowed the training rather than speeding it. I...

bigyaan

AssertionError: each element of save_at [torch.Tensor] must be contained in t_span [torch.Tensor] once and only once, why am I getting this error?

5

My code was running fine for older version of torchdyn but I get the above error in the newer version. What has changed in this newer version?

bigyaan

Passing in multiple arguments

4

**Additional Description** I have a network I wish to train `f(x, x_dot, theta)` where `x` and `x_dot` are the inputs, `theta` are the network weights. This is a slightly odd...

varunagrawal

New Interpolators and Orthogonal Spline Collocation

1

Add new interpolators and the [Orthogonal Spline Collocation](https://www.sciencedirect.com/science/article/pii/S0377042700005094) under [`torchdyn/numerics/interpolators.py`](https://github.com/DiffEqML/torchdyn/blob/master/torchdyn/numerics/interpolators.py). Would be a fine addition to our collection to add a tutorial as well for the interpolators - in preparation...

fedebotu

Implicit Euler method is not compatible with neural ode training

3

First of all, thank you for this amazing library! For my own research, I want to test implicit integrators with adjoint sensitivity method to train a neural network. While implicit...

cyx96

How to get memory usage for "adjoint" and "autograd" method?

1

Thanks for this amazing package! I was trying to test the memory usage of adjoint, as claimed by authors of the original neural ODE paper, the memory usage of adjoint...

cyx96