This issue will be used to track common interface option handling. This is a constant time factor so it's not the biggest deal, but I feel that we can improve some applications by reducing common latency here.

A biexponential synapse, as per the Brian 2 documentation, should implement a current/conductance as g(t)=(tau2/(tau2-tau1))*(exp(-t/tau1)-exp(-t/tau2)).

When playing around with the biexponential synapse concept, I found that maybe the above description might need to be updated. Consider

import matplotlib as mpl
mpl.use('Agg')
import pylab as pl

import numpy as np
from b

Issue:
NeuralODE when driven by a batch input will output an array in which the last index is the time index not the batch index. This is different than the standard convention in all other ML libraries including Flux, TensorFlow, Pytorch, and Chainer.
In Flux.conv the batch output dimensions are
( Data-dimension, channel index, batch index)

the time series output from a NeuralODE is anala

It would be helpful to have some text on how contributions should be made.
This includes how to setup the development environment and how to run the tests.

I'm attempting to run the example posted on the GitHub page for diffeqpy. I've installed diffeqpy successfully on a Mac Pro (2019) running Mac OS X 10.15.4 (Catalina). I'm running the Anaconda distribution for Python ver. 3.7.6.

The code I'm running is the following:

#!/Users/user/opt/anaconda3/bin/python
from diffeqpy import de

def f(u,p,t):
    return -u

u0 = 0.5
tspan = (0.,

We should have a setup so that way the DAE initialization algorithm's default tolerances match those from solve. That would require a setup for delayed tolerance changes.

The current documentation can be misleading regarding which features are ready vs in-progress. I'm not sure if there's a standard practice for this, but if we can summarize the current state in this issue I'll update the README.

Relatedly, we might consider adding a header to the README that warns users that this library is still a WIP, and not to depend on it for anything critical. Again, not

I tend to think that beta1 and beta2 are off. I think that beta1=0.4 and beta2=0.1 may be better defaults than what we have right now, but this would need extensive testing. That's a very high gain but right now we rarely ever reject, and are usually 10-100x below the error estimate we are aiming for.

How can I plot the graphs of the examples of "Solving parametric families of PDEs" and "Solving PDEs with trainable coefficients"? Can you release the full scripts?

What I've noticed is that most (I didn't check all of them) PDE notebooks need to be rerun: all cells just display error messages.

E.g.:

http://benchmarks.juliadiffeq.org/html/MOLPDE/allen_cahn_spectral_wpd.html

http://benchmarks.juliadiffeq.org/html/MOLPDE/kdv_spectral_wpd.html