snrv.Snrv.get_transform_Jacobian

Snrv.get_transform_Jacobian(data)

compute Jacobian of self.transform computational graph output (psi) wrt input (data)

• data is n x dim_in tensor – n instances of dim_in = self.input_size vectors at which to compute Jacobian

• psi = self.transform(data) is n x dim_out – n dim_out = n_comp = no. basis functions in ANN = self.output_size projections of data vectors

• Jacobian[n,i,j] = d(psi_i)/d(data_j) @ data[n,:]

Parameters

data (torch.tensor, n x dim_in, n = observations, dim_in = dimensionality of trajectory) – featurization = self.input_size

Returns

Jacobian – dim_out = dimensionality of output

Return type

torch.tensor, n x dim_out x dim_in, n = observations, dim_in = dimensionality of input,