Scikit-learn
human-activity-recognition
wearable
wrist
time-series
cpu
scikit-learn
WISP / src /wisp_fast /selection.py
Zipeng365's picture
wisp-har 0.2.0: accelerated implementation with speed (exact/fast) and HMM sequence (subject/recording) options
ac7f7cb verified
Raw History Blame Contribute Delete
1.21 kB
"""Split checks and the fit call shared by the accelerated search."""
import numpy as np
def check_splits(*parts):
"""Require sequence metadata, one global vocabulary and disjoint group IDs."""
first = parts[0]
if not first.label_names:
raise ValueError("Supply the ordered global label_names")
seen = set()
for data in parts:
if data.label_names != first.label_names or data.X.shape[1:] != first.X.shape[1:]:
raise ValueError("Split vocabularies and time/channel shapes must match")
if data.subject is None or data.time_index is None or not len(data.y):
raise ValueError("Every split needs nonempty windows, subject and time_index")
if np.min(data.y) < 0 or np.max(data.y) >= len(first.label_names):
raise ValueError("Class ID outside supplied global vocabulary")
groups = set(np.asarray(data.subject).astype(str))
if groups & seen:
raise ValueError("Participant/sequence groups overlap between splits")
seen |= groups
def fit_on(model, data):
return model.fit(data.X, data.y, groups=data.subject, time_index=data.time_index,
n_classes=data.n_classes)