NumPy Shapes, Axes and Filtering

NumPy’s ndarray stores numerical data with a shape and dtype. For this lesson, rows represent three days and columns represent two sensors. The array [[12,18],[14,22],[16,20]] has shape (3,2), two dimensions and six items. Shape is not the same as the total item count. numpy.ndim(array), numpy.shape(array) and numpy.size(array) answer different questions.

A dtype defines how items are represented. Fixed-width integer arrays can overflow where an ordinary Python int could grow, and floating arrays can round. Use an explicit suitable dtype when the task needs predictable representation; do not assume the same letter code means Python’s unlimited integer or a Boolean. Converting a floating array to an integer dtype can discard fractional information.

Arithmetic between same-shaped numerical arrays is normally elementwise. Adding 2 to every reading also uses NumPy’s broadcasting rules for a scalar. A reduction along axis 0 combines rows and returns one result per column, giving sensor means [14,20]. Along axis 1, it combines columns and returns daily means [15,18,18]. Without an axis, numpy.mean averages all six values, giving 17.

The expression readings>=20 creates a Boolean array of the same shape. Indexing readings with that mask returns the selected values [22,20] as a one-dimensional result. It does not retain a rectangular table with empty cells. Use parentheses and elementwise operators for compound array conditions; Python’s scalar and/or operators are not interchangeable with elementwise Boolean array operations.

import numpy as np

readings = np.array([[12, 18], [14, 22], [16, 20]], dtype=np.int64)
print(np.ndim(readings), np.shape(readings), np.size(readings))
print(np.mean(readings, axis=0).tolist())
print(np.mean(readings, axis=1).tolist())
print(float(np.mean(readings)))
print(readings[readings >= 20].tolist())
print(np.reshape(readings, (2, 3)).tolist())

Reshaping reinterprets item positions in a new compatible shape; it does not calculate new readings or prove that the new rows have meaningful labels. Six items can form (2,3), but not (4,2). ravel produces a flattened one-dimensional representation; the result’s copy/view behaviour depends on the array, so make an explicit copy when independent mutation is required.

For numeric text files, specify a delimiter, header handling and dtype when loading. When saving, choose a format that preserves the required precision: fmt="%d" is suitable for this integer dataset, while an inadequate floating format could round information away. The StringIO example keeps the exercise in memory. On a real file, use the supplied task path and avoid overwriting the source during a cleaning exercise.

from io import StringIO
import numpy as np

source = StringIO('left,right\n12,18\n14,22\n16,20\n')
readings = np.loadtxt(source, delimiter=',', skiprows=1, dtype=np.int64)
saved = StringIO()
np.savetxt(saved, readings, delimiter=',', fmt='%d')
saved.seek(0)
restored = np.loadtxt(saved, delimiter=',', dtype=np.int64)
print(np.array_equal(readings, restored))
print(restored.tolist())

Attempt the task before opening its worked response. Record each change of state, explain the rule used, and check the result against the original requirements. Code examples labelled Python 3 use that language’s conventions; pseudocode states its own assumptions.