"""

Here are some notes and frequently used numpy functions.

"""

# creating arrays

np.zeros()

np.ones()

np.zeros_like()

np.ones_like()

np.array()

# these are useful when we need to adjust the dimensions.

np.expand_dims(x, axis=0)

np.squeeze()

np.reshape(x, (1,2,3))

np.transpose(x, (1,2,0)) # permute axis.

#np.swapaxes(x, 0, 1) # this could be done by np.transpose().

# join/split arrays TODO

np.stack()

np.split()

# tile arrays

np.tile()

# serialization/deserialization TODO

# broadcast rules. https://numpy.org/devdocs/user/theory.broadcasting.html

# a numpy arithmetic operation, doesn't require the arrays to have exact shapes.

# smaller arrays could be broadcasted to fit the larger array.

# A's shape: (a1,a2,a3,a4), B's shape: (b1,b2,b3)

# the trailing axes are determined by aligning shapes to the right.

# a1 x a2 x a3 x a4

# b1 X b2 X b3

# operation works if the trailing axes have identical shapes, or on these dimensions one of the two is 1.

# Example #1

# A: 256 x 256 x 3

# B: 3

# Example #2

# A: 8 x 1 x 6 x 1

# B: 7 x 1 x 5

# to fill up the missing dimensions, the smaller will just copy itself along those dimensions.

# this is quite useful for vectorization and sppedup the computation in C instead of in Python.

# it can also be used to perform outer operations.

# element-wise multiplication (when shapes are the same. when it's not there is broadcast.)

x * y

# multiplications

np.dot()

np.matmul()