The CTM TA pattern in rectangular input/kernel shape

[Baizhou-713]

Hello.

I found that something was wrong in the Convolutional Tsetlin Machine. When the CTM deals with rectangular input/kernel, it uses a transposed kernel for convolution (input also has some problems when it is not square). This leads to an error in the ta pattern when the dimension of the transposed kernel is greater than the input data. The correct TA order can only be extracted when the input and kernel are square.

At the same time, I found that in tools.py and Tools.c, the processing order for inputs is "y_pos, x_pos, p_z, p_x, p_y". It is different from the actual test results, which are shown in my file. The actual test results show that the order of ta should be: "x_pos, y_pos, p_y, p_x" (I did not test p_z).

I have attached the test files, hoping they will be helpful to you. Thank you.
model_test.zip