TopolearnFiltering with masks replaces if-statements inside loops and is the basis of pandas filtering.
A comparison like a > 0 gives a boolean array of the same shape. Use it as an index and you get only the elements where the mask is True: a[a > 0]. This is boolean indexing (or masking).
The result is always a 1-D copy of the selected elements, in row-major order, even when a is 2-D. Since it's a copy, modifying the result does not touch a. But assigning through the mask does: a[a < 0] = 0 writes zeros into a in place. That pattern, clipping or replacing values by condition, is extremely common.
Build compound conditions with & (and), | (or), ~ (not), and wrap each comparison in parentheses because these operators bind tighter than < or ==.
Because True counts as 1, masks double as counters:
(a > 0).sum()counts matches;(a > 0).mean()gives the fractionmask.any()andmask.all()answer "is there any / are all"np.nonzero(mask)(ornp.flatnonzero) returns the positions
For "choose between two values depending on a condition", use np.where(cond, x, y), which returns a new array with x where the condition holds and y elsewhere. With a single argument, np.where(cond) returns indices, like np.nonzero.
A mask can also select along one axis: m[m[:, 0] > 5] keeps the rows whose first column exceeds 5, and m[:, mask] keeps columns. The mask length must match that axis.
Example
import numpy as np
t = np.array([12.5, -3.0, 7.1, -0.4, 22.0])
print(t[t > 0]) # [12.5 7.1 22. ]
print((t < 0).sum()) # 2
print((t > 10).mean()) # 0.4
t[t < 0] = 0 # assign through the mask
print(t) # [12.5 0. 7.1 0. 22. ]
x = np.arange(10)
print(x[(x % 2 == 0) & ~(x > 6)]) # [0 2 4 6]
print(np.where(x < 5, "low", "high")) # ['low' ... 'high']
rows = np.array([[1, 9], [6, 2], [8, 3]])
print(rows[rows[:, 0] > 5]) # [[6 2] [8 3]] rows whose col 0 > 5
print(np.nonzero(rows[:, 0] > 5)[0]) # [1 2]Remember
a[mask]returns a 1-D copy of the selected elementsa[mask] = valuemodifies the original in place- Combine conditions with
& | ~and parentheses np.where(cond, x, y)picks element-wise between two options
Watch out for
- Expecting
a[mask]to keep the 2-D shape: it always flattens the selection.