pyharbinger 0.2.0__tar.gz → 0.3.0__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.5
2
2
  Name: pyharbinger
3
- Version: 0.2.0
3
+ Version: 0.3.0
4
4
  Summary: A unified time series event detection framework (Python port of the R harbinger framework)
5
5
  Project-URL: Homepage, https://github.com/cefet-rj-dal/pyharbinger
6
6
  Project-URL: Repository, https://github.com/cefet-rj-dal/pyharbinger
@@ -93,7 +93,7 @@ Every detector extends `Harbinger` and returns a detection table with columns
93
93
  |---|---|
94
94
  | **Core** | `Harbinger` (base contract), deviation / filter / candidate-selection utilities |
95
95
  | **Evaluation** | `HarEval` (hard), `HarEvalSoft` (SoftED, soft matching via the Hungarian algorithm) |
96
- | **Anomaly** | `HanrMl` (forecasting residuals, wraps a pytspredit regressor), `HanrFft`, `HanrHistogram`, `HanrFbiad` |
96
+ | **Anomaly** | `HanrMl` (forecasting residuals, wraps a pytspredit regressor), `HanrFft`, `HanrFftAmoc`, `HanrFftBinseg`, `HanrFftAmocCusum`, `HanrFftBinsegCusum`, `HanrHistogram`, `HanrFbiad` |
97
97
  | **Change point** | `HcpAmoc`, `HcpBinseg`, `HcpPelt` (via `ruptures`), `HcpKswin`, `HcpPageHinkley` |
98
98
  | **Clustering** | `HanctKmeans` (anomalies / discords) |
99
99
  | **Classification** | `HancMl` (supervised; wraps a pydaltoolbox classifier) |
@@ -57,7 +57,7 @@ Every detector extends `Harbinger` and returns a detection table with columns
57
57
  |---|---|
58
58
  | **Core** | `Harbinger` (base contract), deviation / filter / candidate-selection utilities |
59
59
  | **Evaluation** | `HarEval` (hard), `HarEvalSoft` (SoftED, soft matching via the Hungarian algorithm) |
60
- | **Anomaly** | `HanrMl` (forecasting residuals, wraps a pytspredit regressor), `HanrFft`, `HanrHistogram`, `HanrFbiad` |
60
+ | **Anomaly** | `HanrMl` (forecasting residuals, wraps a pytspredit regressor), `HanrFft`, `HanrFftAmoc`, `HanrFftBinseg`, `HanrFftAmocCusum`, `HanrFftBinsegCusum`, `HanrHistogram`, `HanrFbiad` |
61
61
  | **Change point** | `HcpAmoc`, `HcpBinseg`, `HcpPelt` (via `ruptures`), `HcpKswin`, `HcpPageHinkley` |
62
62
  | **Clustering** | `HanctKmeans` (anomalies / discords) |
63
63
  | **Classification** | `HancMl` (supervised; wraps a pydaltoolbox classifier) |
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
4
4
 
5
5
  [project]
6
6
  name = "pyharbinger"
7
- version = "0.2.0"
7
+ version = "0.3.0"
8
8
  description = "A unified time series event detection framework (Python port of the R harbinger framework)"
9
9
  readme = "README.md"
10
10
  requires-python = ">=3.9"
@@ -10,7 +10,16 @@ scikit-learn / scipy / ruptures with thin adapters.
10
10
  from __future__ import annotations
11
11
 
12
12
  from . import datasets, utils
13
- from .anomaly import HanrFbiad, HanrFft, HanrHistogram, HanrMl
13
+ from .anomaly import (
14
+ HanrFbiad,
15
+ HanrFft,
16
+ HanrFftAmoc,
17
+ HanrFftAmocCusum,
18
+ HanrFftBinseg,
19
+ HanrFftBinsegCusum,
20
+ HanrHistogram,
21
+ HanrMl,
22
+ )
14
23
  from .changepoint import HcpAmoc, HcpBinseg, HcpKswin, HcpPageHinkley, HcpPelt
15
24
  from .classification import HancMl
16
25
  from .clustering import HanctKmeans
@@ -30,15 +39,20 @@ from .utils import (
30
39
  har_filter_mad,
31
40
  har_filter_none,
32
41
  har_filter_ratio,
42
+ mas,
33
43
  )
34
44
 
35
- __version__ = "0.2.0"
45
+ __version__ = "0.3.0"
36
46
 
37
47
  __all__ = [
38
48
  "HancMl",
39
49
  "HanctKmeans",
40
50
  "HanrFbiad",
41
51
  "HanrFft",
52
+ "HanrFftAmoc",
53
+ "HanrFftAmocCusum",
54
+ "HanrFftBinseg",
55
+ "HanrFftBinsegCusum",
42
56
  "HanrHistogram",
43
57
  "HanrMl",
44
58
  "HarEval",
@@ -63,5 +77,6 @@ __all__ = [
63
77
  "har_filter_mad",
64
78
  "har_filter_none",
65
79
  "har_filter_ratio",
80
+ "mas",
66
81
  "utils",
67
82
  ]
@@ -15,12 +15,25 @@ All of them reuse the shared :class:`~pyharbinger.harbinger.Harbinger` hooks
15
15
  """
16
16
  from __future__ import annotations
17
17
 
18
+ from math import log
19
+
18
20
  import numpy as np
21
+ import ruptures as rpt
19
22
  from pytspredit import ts_data, ts_projection
20
23
 
21
24
  from .harbinger import Harbinger
25
+ from .utils import har_candidate_selection_highgroup
22
26
 
23
- __all__ = ["HanrFbiad", "HanrFft", "HanrHistogram", "HanrMl"]
27
+ __all__ = [
28
+ "HanrFbiad",
29
+ "HanrFft",
30
+ "HanrFftAmoc",
31
+ "HanrFftAmocCusum",
32
+ "HanrFftBinseg",
33
+ "HanrFftBinsegCusum",
34
+ "HanrHistogram",
35
+ "HanrMl",
36
+ ]
24
37
 
25
38
 
26
39
  # --- hanr_ml ----------------------------------------------------------------
@@ -110,6 +123,98 @@ class HanrFft(Harbinger):
110
123
  return self._restore_refs(anomalies=selection.mask, res=res, threshold=selection.threshold)
111
124
 
112
125
 
126
+ # --- hanr_fft_* (spectral change-point cutoff) ------------------------------
127
+
128
+
129
+ def _spectrum_cut_index(freqs: np.ndarray, binseg: bool) -> int:
130
+ """Cutoff index from a change point in the power spectrum (R ``compute_cut_index``).
131
+
132
+ R uses ``changepoint::cpt.mean`` (change in mean) with method AMOC or BinSeg
133
+ and returns the *last* detected change point. Here ``ruptures`` reproduces it
134
+ with the L2 (mean-shift) cost: AMOC = a single breakpoint; BinSeg = a
135
+ ``3*log(n)`` penalty (matching :class:`HcpPelt`'s calibration and R's cpts).
136
+ """
137
+ k = len(freqs)
138
+ signal = np.asarray(freqs, dtype=float)
139
+ if binseg:
140
+ bkps = rpt.Binseg(model="l2", min_size=1, jump=1).fit(signal).predict(pen=3.0 * log(k))
141
+ else:
142
+ bkps = rpt.Dynp(model="l2", min_size=1, jump=1).fit(signal).predict(n_bkps=1)
143
+ cpts = [b for b in bkps if b < k]
144
+ return cpts[-1] if cpts else 1
145
+
146
+
147
+ class _HanrFftCut(Harbinger):
148
+ """Shared machinery for the FFT high-pass detectors with a spectral cutoff.
149
+
150
+ Subclasses set :attr:`_binseg` (AMOC vs. BinSeg change point) and
151
+ :attr:`_cusum` (whether the cutoff is found on the cumulative spectrum).
152
+ """
153
+
154
+ _binseg = False
155
+ _cusum = False
156
+
157
+ def detect(self, serie, **kwargs):
158
+ if serie is None:
159
+ raise ValueError("No data was provided for computation")
160
+ self._store_refs(serie)
161
+
162
+ fft_signal = np.fft.fft(self.serie)
163
+ spectrum = np.abs(fft_signal) ** 2
164
+ half_spectrum = spectrum[: len(self.serie) // 2 + 1]
165
+
166
+ freqs = np.cumsum(half_spectrum) if self._cusum else half_spectrum
167
+ cutindex = _spectrum_cut_index(freqs, self._binseg)
168
+ n = len(fft_signal)
169
+
170
+ # Zero out low frequencies (high-pass), both symmetric halves.
171
+ fft_signal[:cutindex] = 0
172
+ fft_signal[n - cutindex - 1 :] = 0
173
+
174
+ filtered_series = np.real(np.fft.ifft(fft_signal))
175
+
176
+ res = self.har_distance(filtered_series)
177
+ outliers = self.har_outliers(res)
178
+ selection = self.har_outliers_check(outliers, res, self.serie)
179
+
180
+ return self._restore_refs(anomalies=selection.mask, res=res, threshold=selection.threshold)
181
+
182
+
183
+ class HanrFftAmoc(_HanrFftCut):
184
+ """FFT high-pass detector; cutoff = AMOC change point of the power spectrum."""
185
+
186
+ _binseg = False
187
+ _cusum = False
188
+
189
+
190
+ class HanrFftBinseg(_HanrFftCut):
191
+ """FFT high-pass detector; cutoff = last BinSeg change point of the power spectrum."""
192
+
193
+ _binseg = True
194
+ _cusum = False
195
+
196
+
197
+ class HanrFftAmocCusum(_HanrFftCut):
198
+ """FFT high-pass detector; cutoff = AMOC change point of the cumulative spectrum."""
199
+
200
+ _binseg = False
201
+ _cusum = True
202
+
203
+
204
+ class HanrFftBinsegCusum(_HanrFftCut):
205
+ """FFT high-pass detector; cutoff = last BinSeg change point of the cumulative spectrum.
206
+
207
+ Uses the high-group candidate selection (as in R ``hanr_fft_binseg_cusum``).
208
+ """
209
+
210
+ _binseg = True
211
+ _cusum = True
212
+
213
+ def __init__(self):
214
+ super().__init__()
215
+ self.har_outliers_check = har_candidate_selection_highgroup
216
+
217
+
113
218
  # --- hanr_histogram ---------------------------------------------------------
114
219
 
115
220
 
@@ -38,9 +38,24 @@ __all__ = [
38
38
  "har_filter_mad",
39
39
  "har_filter_none",
40
40
  "har_filter_ratio",
41
+ "mas",
41
42
  ]
42
43
 
43
44
 
45
+ def mas(x, order: int) -> np.ndarray:
46
+ """Simple moving-average smoother (Python port of R ``mas``).
47
+
48
+ Returns the length-``(len(x) - order + 1)`` vector of window means
49
+ ``mean(x[t:t+order])`` for ``t = 0 .. len(x)-order``.
50
+ """
51
+ x = np.asarray(x, dtype=float)
52
+ n = len(x)
53
+ if order < 1 or order > n:
54
+ raise ValueError("order must be between 1 and len(x).")
55
+ cumulative = np.concatenate(([0.0], np.cumsum(x)))
56
+ return (cumulative[order:] - cumulative[:-order]) / order
57
+
58
+
44
59
  @dataclass
45
60
  class FilterResult:
46
61
  """Candidate outlier positions and the (inf, sup) threshold that produced them.
File without changes
File without changes