I have two numpy arrays of same shape (3, 3, 3, 64) - x & y. I then compute the pth percentile of the two arrays as their respective threshold which is then to be used to remove all elements less than the threshold. The code I have is as follows:
The aim is to remove all values in 'x' based on the condition: all values in 'x' < x_threshold AND all all values in 'y' < y_threshold:
# Create two np arrays of same dimension-
x = np.random.rand(3, 3, 3, 64)
y = np.random.rand(3, 3, 3, 64)
x.shape, y.shape
# ((3, 3, 3, 64), (3, 3, 3, 64))
x.min(), x.max()
# (0.0003979483351387314, 0.9995167558342761)
y.min(), y.max()
# (0.0006328536816179176, 0.9999504057216633)
# Compute 20th percentile as threshold for both np arrays-
x_threshold = np.percentile(x, 20)
y_threshold = np.percentile(y, 20)
print(f"x_threshold = {x_threshold:.4f} & y_threshold = {y_threshold:.4f}")
# x_threshold = 0.2256 & y_threshold = 0.1958
x[x < x_threshold].shape, y[y < y_threshold].shape
# ((346,), (346,))
# For 'x' try and remove all elements which for both of these conditions are true-
x[x < x_threshold and y < y_threshold]
The last line of code throws the error:
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()
Solutions?
I came with the following code as a possible solution:
x_index = x < x_threshold
y_index = y < y_threshold
x_index.shape, y_index.shape
# ((3, 3, 3, 64), (3, 3, 3, 64))
x_index == y_index
# Find indices where the 2 conditions are NOT satisfied-
x[np.where(x_index != y_index)].shape
# (542,)
# Find indices where the 2 conditions are satisfied-
x[np.where(x_index == y_index)].shape
# (1186,)
# Create a copy of 'x'-
x_mod = x.copy()
# Set all values in 'x_mod' where the 2 conditions are satisfied-
x_mod[np.where(x_index == y_index)] = 0
# Sanity check-
np.count_nonzero(x_mod)
# 542
The count of 542 matches and thereby shows that the code works. Of course, the solution presented by user "bb1" below is more elegant and just a one-liner!
