49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164 | @dataclass
class HypothesisEngine:
"""Emits candidate hypotheses from numeric observation series."""
minimum_confidence: float = 0.4
generated: list[Hypothesis] = field(default_factory=list)
def from_series(
self,
name: str,
values: object,
domain: str = Domain.GENERAL,
) -> list[Hypothesis]:
"""Scan one series for trend, periodicity and outlier patterns."""
data_values = np.asarray(values, dtype=float).ravel()
if data_values.size < 4:
msg = "at least four observations are required"
raise ValueError(msg)
found: list[Hypothesis] = []
found.extend(self._trend_hypotheses(name, data_values, domain))
found.extend(self._periodicity_hypothesis(name, data_values, domain))
found.extend(self._outlier_hypothesis(name, data_values, domain))
kept = [item for item in found if item.confidence >= self.minimum_confidence]
self.generated.extend(kept)
return kept
def from_pair(
self,
first_name: str,
first_values: object,
second_name: str,
second_values: object,
domain: str = Domain.GENERAL,
) -> list[Hypothesis]:
"""Emit a correlation hypothesis between two aligned series."""
first_array = np.asarray(first_values, dtype=float).ravel()
second_array = np.asarray(second_values, dtype=float).ravel()
if first_array.size != second_array.size or first_array.size < 4:
msg = "series must be aligned and contain at least four points"
raise ValueError(msg)
correlation = pearson_correlation(first_array, second_array)
strength = abs(correlation.r)
if strength < 0.5:
return []
direction = "drives" if correlation.r > 0 else "suppresses"
confidence = min(0.95, strength)
hypothesis = Hypothesis(
statement=f"{first_name} {direction} {second_name} (pearson r={correlation.r:+.3f})",
domain=domain,
confidence=confidence,
rationale="linear correlation strength above 0.5",
)
if confidence >= self.minimum_confidence:
self.generated.append(hypothesis)
return [hypothesis]
def _trend_hypotheses(self, name: str, values: np.ndarray, domain: str) -> list[Hypothesis]:
indices = np.arange(values.size, dtype=float)
slope, intercept = np.polyfit(indices, values, 1)
fitted = slope * indices + intercept
residual_std = float(np.std(values - fitted, ddof=1)) or 1e-12
signal = abs(slope) * values.size / (residual_std * 4.0)
confidence = float(np.clip(signal / (1.0 + signal), 0.0, 0.95))
if confidence < self.minimum_confidence:
return []
direction = "rising" if slope > 0 else "falling"
return [
Hypothesis(
statement=f"{name} shows a {direction} linear trend (slope={slope:.4g} per step)",
domain=domain,
confidence=confidence,
rationale="linear fit slope dominates residual noise",
)
]
def _periodicity_hypothesis(
self, name: str, values: np.ndarray, domain: str
) -> list[Hypothesis]:
centered = values - values.mean()
if float(np.std(centered)) == 0.0:
return []
correlations = np.correlate(centered, centered, mode="full")[values.size - 1 :]
correlations /= correlations[0] if correlations[0] != 0 else 1.0
interior = correlations[1:]
best_lag = int(np.argmax(np.abs(interior))) + 1
peak = abs(float(correlations[best_lag]))
confidence = float(np.clip(peak * 0.9, 0.0, 0.9))
if confidence < self.minimum_confidence or best_lag == values.size - 1:
return []
return [
Hypothesis(
statement=f"{name} repeats with a period of about "
f"{best_lag} observations (autocorrelation {correlations[best_lag]:+.3f})",
domain=domain,
confidence=confidence,
rationale="autocorrelation peak at non-zero lag",
)
]
def _outlier_hypothesis(self, name: str, values: np.ndarray, domain: str) -> list[Hypothesis]:
median = float(np.median(values))
mad = float(np.median(np.abs(values - median))) or 1e-12
robust_z = np.abs(values - median) / (1.4826 * mad)
outlier_count = int(np.count_nonzero(robust_z > 3.5))
if outlier_count == 0:
return []
confidence = float(np.clip(0.4 + 0.1 * outlier_count, 0.0, 0.85))
return [
Hypothesis(
statement=f"{name} contains {outlier_count} anomalous "
"observation(s) worth investigating",
domain=domain,
confidence=confidence,
rationale="MAD-robust z-score above 3.5",
)
]
|