File size: 2,437 Bytes
04ae9bb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
"""Process features from arrived, label-free test events only.

This module has no clinical-table input or cohort lookup. Learned model and
preprocessing parameters may be loaded separately by the inference process.
"""
from __future__ import annotations

import numpy as np


class StreamProcessFeatures:
    FIELDS = {"visit_id", "patient_id", "visit_month"}

    def __init__(self, months, mode="observed"):
        if mode not in {"observed", "current_only"}:
            raise ValueError(mode)
        self.months = [int(m) for m in months]
        self.mode = mode
        self.history = {}
        self.last_month = -1
        self.seen_visits = set()

    @property
    def names(self):
        return [f"v{m}" for m in self.months] + ["control", "gap", "prior_count", "visit_month"]

    def update(self, event):
        if set(event) != self.FIELDS:
            raise ValueError("Inference events must contain exactly visit_id, patient_id, visit_month")
        patient, month = int(event["patient_id"]), int(event["visit_month"])
        visit = str(event["visit_id"])
        if month < self.last_month:
            raise ValueError("Test events must arrive in chronological order")
        if visit in self.seen_visits:
            raise ValueError("Duplicate test event")
        self.last_month = month
        self.seen_visits.add(visit)
        if self.mode == "current_only":
            flags = [1.0 if m == month else -1.0 for m in self.months]
            return np.asarray(flags + [-1., -1., -1., float(month)], dtype=np.float32)

        previous = self.history.get(patient, set())
        past = sorted(m for m in previous if m < month)
        available = previous | {month}
        self.history[patient] = available
        flags = [-1. if m > month else float(m in available) for m in self.months]
        first_nonzero = min((m for m in available if m > 0), default=None)
        control = -1. if month < 12 else float(first_nonzero == 12)
        gap = float(month - past[-1]) if past else -1.
        return np.asarray(flags + [control, gap, float(len(past)), float(month)], dtype=np.float32)


def features_for_events(events, months, mode="observed"):
    builder = StreamProcessFeatures(months, mode)
    rows, visits = [], []
    for event in events:
        rows.append(builder.update(event))
        visits.append(event["visit_id"])
    return np.asarray(rows, dtype=np.float32), visits, builder.names