olliepugh/parkr

A lightweight basic neural networks library in Python

parkr's Issues

Fatal bug when trying to export training results

When trying to export training without providing validation data a fatal error occurs.

Line 277 in train.export_file.write("f{training_loss}, {validation_loss}\n")
UnboundLocalError: local variable 'validation_loss' referenced before assignment

Investigate random installs in requirements.txt

Not sure why things such as

Pillow
matplotlib

are in the requirements.txt

Fatal error on more than one output node

Fatal error when trying to train with more than one output node.

(env) C:\Users\olive\OneDrive\Programming\Python\Parkr>py main.py
Traceback (most recent call last):
  File "C:\Users\olive\OneDrive\Programming\Python\Parkr\main.py", line 16, in <module>
    my_network.train(epochs, (training_data, expected_data))
  File "C:\Users\olive\OneDrive\Programming\Python\Parkr\src\parkr\network.py", line 180, in train
    row_weight_change, row_bias_change = self.__generate_changes(activated_forward_pass, expected_batches[batch_count][i], step_size, prev_changes=(prev_weight_change, prev_bias_change), **kwargs)  # generate the changes required        
  File "C:\Users\olive\OneDrive\Programming\Python\Parkr\src\parkr\network.py", line 251, in __generate_changes
    delta = np.dot((expected-activated_result[i]), derived)
  File "<__array_function__ internals>", line 5, in dot
ValueError: shapes (1,2) and (1,2) not aligned: 2 (dim 1) != 1 (dim 0)