-
Notifications
You must be signed in to change notification settings - Fork 1
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
move statistics stuff into its own file. add online statistics comput…
…ation
- Loading branch information
Showing
7 changed files
with
493 additions
and
321 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
292 changes: 292 additions & 0 deletions
292
mondey_backend/src/mondey_backend/routers/statistics.py
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,292 @@ | ||
| from __future__ import annotations | ||
|
|
||
| import datetime | ||
| from collections.abc import Sequence | ||
|
|
||
| import numpy as np | ||
| from sqlmodel import col | ||
| from sqlmodel import select | ||
|
|
||
| from ..dependencies import SessionDep | ||
| from ..models.milestones import Milestone | ||
| from ..models.milestones import MilestoneAgeScore | ||
| from ..models.milestones import MilestoneAgeScoreCollection | ||
| from ..models.milestones import MilestoneAnswer | ||
| from ..models.milestones import MilestoneAnswerSession | ||
| from ..models.milestones import MilestoneGroup | ||
| from ..models.milestones import MilestoneGroupAgeScore | ||
| from ..models.milestones import MilestoneGroupAgeScoreCollection | ||
| from .utils import _get_answer_session_child_ages_in_months | ||
| from .utils import _get_expected_age_from_scores | ||
| from .utils import add | ||
|
|
||
|
|
||
| # see: https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance | ||
| # reason for not using existing package: bessel correction usually not respected | ||
| # we are using Welford's method here. This necessitates recording the count | ||
| def _add_sample( | ||
| count: int, | ||
| mean: float, | ||
| m2: float, | ||
| new_value: float, | ||
| ) -> tuple[int, float, float]: | ||
| count += 1 | ||
| delta = new_value - mean | ||
| mean += delta / count | ||
| delta2 = new_value - mean | ||
| m2 += delta * delta2 | ||
| return count, mean, m2 | ||
|
|
||
|
|
||
| def _finalize_statistics( | ||
| count: int | np.ndarray[int], | ||
| mean: float | np.ndarray[float], | ||
| m2: float | np.ndarray[float], | ||
| ) -> tuple[int | np.ndarray[int], float | np.ndarray[float], float | np.ndarray[float]]: | ||
| if isinstance(count, int): | ||
| if count < 2: | ||
| return count, mean, 0.0 | ||
| else: | ||
| variance = m2 / (count - 1) | ||
| return count, mean, np.sqrt(variance) | ||
| elif isinstance(count, np.ndarray): | ||
| with np.errstate(invalid="ignore"): | ||
| valid_counts = count >= 2 | ||
| variance = m2 | ||
| variance[valid_counts] /= count[valid_counts] - 1 | ||
| variance[not valid_counts] = 0.0 | ||
| return count, np.nan_to_num(mean), np.nan_to_num(np.sqrt(variance)) | ||
| else: | ||
| raise ValueError("given values must be of type int|float|np.ndarray") | ||
|
|
||
|
|
||
| def _get_statistics_by_age( | ||
| answers: Sequence[MilestoneAnswer], | ||
| child_ages: dict[int, int], | ||
| count: np.ndarray = None, | ||
| avg: np.ndarray = None, | ||
| stddev: np.ndarray = None, | ||
| ) -> tuple[np.ndarray, np.ndarray, np.ndarray]: | ||
| if count is None or avg is None or stddev is None: | ||
| max_age_months = 72 | ||
| count = np.zeros(max_age_months + 1) | ||
| avg = np.zeros(max_age_months + 1) | ||
| stddev = np.zeros(max_age_months + 1) | ||
|
|
||
| # online algorithm computes variance, so need to square stddev first | ||
| var = stddev**2 | ||
|
|
||
| if child_ages == {}: | ||
| return count, avg, stddev | ||
|
|
||
| for answer in answers: | ||
| age = child_ages[answer.answer_session_id] # type: ignore | ||
| new_count, new_avg, new_var = _add_sample( | ||
| count[age], avg[age], var[age], answer.answer + 1 | ||
| ) | ||
| count[age] = new_count | ||
| avg[age] = new_avg | ||
| var[age] = new_var | ||
|
|
||
| count, avg, stddev = _finalize_statistics(count, avg, var) | ||
|
|
||
| return count, avg, stddev | ||
|
|
||
|
|
||
| def calculate_milestone_statistics_by_age( | ||
| session: SessionDep, | ||
| milestone_id: int, | ||
| ) -> MilestoneAgeScoreCollection: | ||
| """ | ||
| _summary_ | ||
| Parameters | ||
| ---------- | ||
| session : SessionDep | ||
| _description_ | ||
| milestone_id : int | ||
| _description_ | ||
| answers : Sequence[MilestoneAnswer] | None, optional | ||
| _description_, by default None | ||
| Returns | ||
| ------- | ||
| MilestoneAgeScoreCollection | ||
| _description_ | ||
| """ | ||
| # get the newest statistics for the milestone | ||
| last_statistics = session.exec( | ||
| select(MilestoneAgeScoreCollection) | ||
| .where(col(MilestoneAgeScoreCollection.milestone_id) == milestone_id) | ||
| .order_by(MilestoneAgeScoreCollection.created_at.desc()) | ||
| ).first() | ||
|
|
||
| # initialize avg and stddev scores with the last known statistics or to None if no statistics are available | ||
| count = None | ||
| avg_scores = None | ||
| stddev_scores = None | ||
| if last_statistics is not None: | ||
| last_scores = last_statistics.scores | ||
| count = np.array([score.count for score in last_scores]) | ||
| avg_scores = np.array([score.avg_score for score in last_scores]) | ||
| stddev_scores = np.array([score.stddev_score for score in last_scores]) | ||
|
|
||
| child_ages = _get_answer_session_child_ages_in_months(session) | ||
|
|
||
| if last_statistics is None: | ||
| answers_query = select(MilestoneAnswer).where( | ||
| MilestoneAnswer.milestone_id == milestone_id | ||
| ) | ||
| else: | ||
| answers_query = ( | ||
| select(MilestoneAnswer) | ||
| .join( | ||
| MilestoneAnswerSession, | ||
| MilestoneAnswer.answer_session_id == MilestoneAnswerSession.id, | ||
| ) | ||
| .where( | ||
| MilestoneAnswer.milestone_id == milestone_id, | ||
| MilestoneAnswerSession.created_at > last_statistics.created_at, | ||
| ) | ||
| ) | ||
| answers = session.exec(answers_query).all() | ||
|
|
||
| count, avg, stddev = _get_statistics_by_age( | ||
| answers, child_ages, count=count, avg=avg_scores, stddev=stddev_scores | ||
| ) | ||
| expected_age = _get_expected_age_from_scores(avg) | ||
| new_id = last_statistics.id + 1 if last_statistics is not None else 1 | ||
| return MilestoneAgeScoreCollection( | ||
| id=new_id, | ||
| milestone_id=milestone_id, | ||
| expected_age=expected_age, | ||
| created_at=datetime.datetime.now(), | ||
| scores=[ | ||
| MilestoneAgeScore( | ||
| collection_id=new_id, | ||
| count=count[age], | ||
| avg_score=avg[age], | ||
| stddev_score=stddev[age], | ||
| age_months=age, | ||
| expected_score=4 | ||
| if age >= expected_age | ||
| else 1, # TODO: placeholder algorithm? how does the model behind this look like really? | ||
| ) | ||
| for age in range(0, len(avg)) | ||
| ], | ||
| ) | ||
|
|
||
|
|
||
| def calculate_milestonegroup_statistics_by_age( | ||
| session: SessionDep, | ||
| milestonegroup_id, | ||
| ) -> MilestoneGroupAgeScoreCollection: | ||
| """ | ||
| _summary_ | ||
| Parameters | ||
| ---------- | ||
| session : SessionDep | ||
| _description_ | ||
| milestonegroup_id : _type_ | ||
| _description_ | ||
| answers : Sequence[MilestoneAnswer] | None, optional | ||
| _description_, by default None | ||
| Returns | ||
| ------- | ||
| MilestoneGroupAgeScoreCollection | ||
| _description_ | ||
| """ | ||
|
|
||
| # get the newest statistics for the milestonegroup | ||
| last_statistics = session.exec( | ||
| select(MilestoneGroupAgeScoreCollection) | ||
| .where( | ||
| col(MilestoneGroupAgeScoreCollection.milestonegroup_id) == milestonegroup_id | ||
| ) | ||
| .order_by(MilestoneGroupAgeScoreCollection.created_at.desc()) | ||
| ).first() | ||
|
|
||
| count = None | ||
| avg_scores = None | ||
| stddev_scores = None | ||
| if last_statistics is None: | ||
| max_age_months = 72 | ||
| count = np.zeros(max_age_months + 1) | ||
| avg_scores = np.zeros(max_age_months + 1) | ||
| stddev_scores = np.zeros(max_age_months + 1) | ||
| else: | ||
| count = np.array([score.count for score in last_statistics.scores]) | ||
| avg_scores = np.array([score.avg_score for score in last_statistics.scores]) | ||
| stddev_scores = np.array( | ||
| [score.stddev_score for score in last_statistics.scores] | ||
| ) | ||
|
|
||
| child_ages = _get_answer_session_child_ages_in_months(session) | ||
|
|
||
| if last_statistics is None: | ||
| answer_query = select(MilestoneAnswer).where( | ||
| col(MilestoneAnswer.milestone_group_id) == milestonegroup_id | ||
| ) | ||
| else: | ||
| answer_query = ( | ||
| select(MilestoneAnswer) | ||
| .join( | ||
| MilestoneAnswerSession, | ||
| MilestoneAnswer.answer_session_id == MilestoneAnswerSession.id, | ||
| ) | ||
| .where( | ||
| MilestoneAnswer.milestone_group_id == milestonegroup_id, | ||
| MilestoneAnswerSession.created_at > last_statistics.created_at, | ||
| ) | ||
| ) | ||
|
|
||
| answers = session.exec(answer_query).all() | ||
|
|
||
| count, avg, stddev = _get_statistics_by_age( | ||
| answers, child_ages, count=count, avg=avg_scores, stddev=stddev_scores | ||
| ) | ||
| new_id = last_statistics.id + 1 if last_statistics is not None else 1 | ||
| return MilestoneGroupAgeScoreCollection( | ||
| id=new_id, | ||
| milestonegroup_id=milestonegroup_id, | ||
| scores=[ | ||
| MilestoneGroupAgeScore( | ||
| collection_id=new_id, | ||
| age_months=age, | ||
| count=count[age], | ||
| avg_score=avg[age], | ||
| stddev_score=stddev[age], | ||
| milestonegroup_id=milestonegroup_id, | ||
| ) | ||
| for age in range(0, len(avg)) | ||
| ], | ||
| created_at=datetime.datetime.now(), | ||
| ) | ||
|
|
||
|
|
||
| def recompute_milestonegroup_statistics(session: SessionDep): | ||
| # fetch all milestonegroup statsitcs and check how old they are. Then | ||
| # recompute the ones that are older than timedelta and put back into database | ||
| # do the same for milestone statistics | ||
|
|
||
| milestonegroups = session.exec(select(MilestoneGroup.id)).all() | ||
| for milestonegroup in milestonegroups: | ||
| statistics = calculate_milestonegroup_statistics_by_age(session, milestonegroup) | ||
| for score in statistics.scores: | ||
| add(session, score) | ||
| add(session, statistics) | ||
|
|
||
|
|
||
| def recompute_milestone_statistics(session: SessionDep): | ||
| # fetch all milestonegroup statsitcs and check how old they are. Then | ||
| # recompute the ones that are older than timedelta and put back into database | ||
| # do the same for milestone statistics | ||
|
|
||
| milestones = session.exec(select(Milestone.id)).all() | ||
| for milestone in milestones: | ||
| statistics = calculate_milestone_statistics_by_age(session, milestone) # type: ignore | ||
| for score in statistics.scores: | ||
| add(session, score) | ||
| add(session, statistics) |
Oops, something went wrong.