Select by Condition

Position is useful when we already know where a value sits. A different question is, “Which readings are at least 22 degrees?” The answer depends on the values, so we begin with a condition rather than a row or column number.

The array has shape (3, 2): three observations by two sensors.

A Comparison Produces a Boolean Array

Compare the whole array with one threshold:

[[False False]
 [False  True]
 [False  True]]
(3, 2)

NumPy performs the comparison at every position. The resulting boolean array, often called a mask, has the same shape as readings. Each True marks a position whose value meets the condition. Here those positions contain 22.0 and 24.0.

The condition includes equality. If the question had said “above 22,” we would use readings > 22.0, and 22.0 would no longer be selected. The operator is part of the data question, not a minor spelling choice.

Use the Mask to Select Values

Place a same-shaped boolean mask inside the brackets:

[22. 24.]
(2,)

NumPy collects the values at True positions into a one-dimensional result. The source was a table, but this question asks for individual readings rather than complete observations. The result therefore has shape (2,), not (3, 2).

Boolean selection produces a result independent of the source array. We will use that fact here, but Lesson 8 will establish the storage rule with direct evidence and compare it with slices and transpose. For now, the important point is the selection rule: one boolean decision corresponds to each source position.

Select Complete Rows with One Stated Rule

Sometimes the question is about observations rather than individual values. Suppose we need every observation where the second sensor reports at least 22.0. First derive one boolean value per observation:

[False  True  True]
(3,)

This mask has shape (3,) because it contains one decision for each observation. Use it on axis 0 to keep complete rows:

[[21.5 22. ]
 [19.  24. ]]
(2, 2)

The condition names the second sensor, but the selection keeps both sensors for every accepted observation. This distinction matters when several arrays store aligned information about the same rows: derive one row mask, then apply that same decision to each aligned array.