Flag the Odd Readings — medium Machine Learning in Code problem | Incognition

Flag the Odd Readings

10 pts · 10 coins

Machine Learning in CodeMedium

Flag the Odd Readings

A greenhouse sensor reports [temperature_c, humidity_percent] every hour. Most hours look alike; a faulty hour sits far away from the rest. You want DBSCAN to mark those hours as noise.

Temperature and humidity are in different units, so scale them first.

Task: write odd_hours(readings, eps, min_samples).

Scale the columns with StandardScaler().fit_transform(...).
Fit DBSCAN(eps=eps, min_samples=min_samples) on the scaled data.
Return the row indices of the readings labelled -1, from smallest to largest, as plain ints.

If nothing is noise, return an empty list.

Examples

readings = [[21.0, 40], [21.4, 42], [20.8, 41], [21.2, 39], [21.1, 43], [35.0, 41], [21.3, 40], [20.9, 70], [21.0, 42], [21.5, 41]]

eps = 0.8

min_samples = 3

returns [5, 7]

readings = [[18.2, 55], [18.5, 54], [18.1, 56], [26.0, 30], [26.3, 31], [25.9, 29], [18.4, 55], [26.1, 30], [22.0, 80], [18.3, 57], [26.2, 32]]

eps = 0.6

min_samples = 3

returns [8]

Hints

solution.py

1from sklearn.cluster import DBSCAN
2from sklearn.preprocessing import StandardScaler
3
4
5def odd_hours(readings, eps, min_samples):
6    pass
7