Scikit-learn
human-activity-recognition
wearable
wrist
time-series
cpu
scikit-learn
File size: 1,437 Bytes
93a9282
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
"""Context for shared feature blocks during a search (``_SEARCH``); ``_ACTIVE`` is never set here."""
from contextvars import ContextVar

_SEARCH = ContextVar("wisp_search_memory", default=None)
_ACTIVE = ContextVar("wisp_feature_store", default=None)


def transform_blocks(blocks, X):
    config = _ACTIVE.get()
    shared_search = _SEARCH.get()
    if shared_search is not None and config is None:
        return shared_search.transform(blocks,X)
    if config is None:
        shared = len(blocks)>=2 and all(getattr(b,"zscore",False) and hasattr(b,"transform_normalized") for b in blocks)
        if shared:
            from wisp.cpu.features import _as_3d, _safe_zscore_per_window
            from . import features as fx
            if X.flags.c_contiguous:
                z=fx._normalize(_as_3d(X))
                return [b.transform_normalized(z) for b in blocks]
            canonical=(fx.RandomConvSketch,fx.IntervalDistributionSketch,fx.WearableFeatureExtractor)
            if fx._time_major_view(X) and all(type(b) in canonical and not b.fast_profile for b in blocks):
                # Keep the original reduction layout; reuse only if the actual
                # normalized buffer can enter all three packed backends.
                z=_safe_zscore_per_window(_as_3d(X))
                if z.flags.c_contiguous:return [b.transform_normalized(z) for b in blocks]
        return [block.transform(X) for block in blocks]