Output Derivatives in Inverse Problem!!!!

[ppandit95]

Dear Lu Lu

Firstly, thanks alot for bringing in this open source library and it works great.....

Secondly,I am trying to solve inverse problem involving simple Fick's First Law of Diffusion wherein I wish to get the values of derivatives at each training point in post processing so that I can verify my simulation and from one of the examples I implemented it as
def get_derivatives(x,M): dM1_x = dde.grad.jacobian(M,x,i=0,j=0) dM2_x = dde.grad.jacobian(M,x,i=1,j=0) return [dM1_x,dM2_x]
After training the model I called these below lines -
f=model.predict(observe_x,operator = get_derivatives) np.savetxt("Derivatives.dat",np.array(f))

But upon running I am getting an output as
Traceback (most recent call last): File "hPINN-Alternative_not_working.py", line 268, in <module> np.savetxt(f"Derivatives.dat",[f[0],f[1]]) File "<__array_function__ internals>", line 180, in savetxt File "/home/cmslab/.local/lib/python3.8/site-packages/numpy/lib/npyio.py", line 1507, in savetxt raise ValueError( ValueError: Expected 1D or 2D array, got 3D array instead
Any help in this regard will prove beneficial.

many Thanks
Pushkar

[tsarikahin]

The problem here is that np.savetxt. Check here https://stackoverflow.com/questions/51218937/saving-a-3d-numpy-array-to-txt-file.
I think this should work

def get_derivatives(x,M): 
      dM1_x = dde.grad.jacobian(M,x,i=0,j=0) 
      dM2_x = dde.grad.jacobian(M,x,i=1,j=0) 
      return dM1_x,dM2_x

f=model.predict(observe_x,operator = get_derivatives)
res_stacked = np.hstack((observe_x.reshape(-1,1),f[0].reshape(-1,1),f[1].reshape(-1,1)))
np.savetxt("Derivatives.dat",res_stacked)

[ppandit95]

Thanks @tsarikahin for helping as I am able to get the derivatives ......