Parameters prediction order
Issue
when trying to infer my 21 parameters for the cattle dataset, it appeared that the order of the target parameters was permuted in the alphabetical order and not the alphanumerical order.
For instance:
# In the csv file:
['p1', 'p2', 'p11']
# In the dataloader in the training loop:
['p1', 'p11', 'p2']
# because 'p11' < 'p2' in alphabetical order
It is not a real problem during the training loop, because the network learns to infer in another order, which is not a problem for the net.
Example
# Alphabetical order of range(21):
input_permutation = np.array([0,1,10,11,12,13,14,15,16,17,18,19,2,20,3,4,5,6,7,8,9])
# permutation to apply to get the index right:
inversed_permutation = np.argsort(index)