Préparation WEI 2023

apps/wei/forms/surveys/__init__.py

@@ -2,11 +2,11 @@
 # SPDX-License-Identifier: GPL-3.0-or-later
 
 from .base import WEISurvey, WEISurveyInformation, WEISurveyAlgorithm
-from .wei2022 import WEISurvey2022
+from .wei2023 import WEISurvey2023
 
 
 __all__ = [
     'WEISurvey', 'WEISurveyInformation', 'WEISurveyAlgorithm', 'CurrentSurvey',
 ]
 
-CurrentSurvey = WEISurvey2022
+CurrentSurvey = WEISurvey2023

apps/wei/forms/surveys/wei2023.py

@@ -0,0 +1,296 @@
+# Copyright (C) 2018-2023 by BDE ENS Paris-Saclay
+# SPDX-License-Identifier: GPL-3.0-or-later
+
+import time
+from functools import lru_cache
+from random import Random
+
+from django import forms
+from django.db import transaction
+from django.db.models import Q
+from django.utils.translation import gettext_lazy as _
+
+from .base import WEISurvey, WEISurveyInformation, WEISurveyAlgorithm, WEIBusInformation
+from ...models import WEIMembership
+
+WORDS = [
+	'ABBA', 'After', 'Alcoolique anonyme', 'Ambiance festive', 'Années 2000', 'Apéro', 'Art',
+	'Baby foot billard biere pong', 'BBQ', 'Before', 'Bière pong', 'Bon enfant', 'Calme', 'Canapé',
+	'Chanson paillarde', 'Chanson populaire', 'Chartreuse', 'Cheerleader', 'Chill', 'Choré',
+	'Cinéma', 'Cocktail', 'Comédie musicle', 'Commercial', 'Copaing', 'Danse', 'Dancefloor',
+	'Electro', 'Fanfare', 'Gin tonic', 'Inclusif', 'Jazz', "Jeux d'alcool", 'Jeux de carte',
+	'Jeux de rôle', 'Jeux de société', 'JUL', 'Jus de fruit', 'Kfet', 'Kleptomanie assurée',
+	'LGBTQ+', 'Livre', 'Morning beer', 'Musique', 'NAPS', 'Paillettes', 'Pastis', 'Paté Hénaff',
+	'Peluche', 'Pena baiona', "Peu d'alcool", 'Pilier de bar', 'PMU', 'Poulpe', 'Punch', 'Rap',
+	'Réveil', 'Rock', 'Rugby', 'Sandwich', 'Serge', 'Shot', 'Sociable', 'Spectacle', 'Techno',
+	'Techno house', 'Thérapie Taxi', 'Tradition kchanaises', 'Troisième mi-temps', 'Turn up',
+	'Vodka', 'Vodka pomme', 'Volley', 'Vomi stratégique'
+]
+
+
+class WEISurveyForm2023(forms.Form):
+    """
+    Survey form for the year 2023.
+    Members choose 20 words, from which we calculate the best associated bus.
+    """
+
+    word = forms.ChoiceField(
+        label=_("Choose a word:"),
+        widget=forms.RadioSelect(),
+    )
+
+    def set_registration(self, registration):
+        """
+        Filter the bus selector with the buses of the current WEI.
+        """
+        information = WEISurveyInformation2023(registration)
+        if not information.seed:
+            information.seed = int(1000 * time.time())
+            information.save(registration)
+            registration._force_save = True
+            registration.save()
+
+        if self.data:
+            self.fields["word"].choices = [(w, w) for w in WORDS]
+            if self.is_valid():
+                return
+
+        rng = Random((information.step + 1) * information.seed)
+
+        words = None
+
+        buses = WEISurveyAlgorithm2023.get_buses()
+        informations = {bus: WEIBusInformation2023(bus) for bus in buses}
+        scores = sum((list(informations[bus].scores.values()) for bus in buses), [])
+        average_score = sum(scores) / len(scores)
+
+        preferred_words = {bus: [word for word in WORDS
+                                 if informations[bus].scores[word] >= average_score]
+                           for bus in buses}
+        while words is None or len(set(words)) != len(words):
+            # Ensure that there is no the same word 2 times
+            words = [rng.choice(words) for _ignored2, words in preferred_words.items()]
+        rng.shuffle(words)
+        words = [(w, w) for w in words]
+        self.fields["word"].choices = words
+
+
+class WEIBusInformation2023(WEIBusInformation):
+    """
+    For each word, the bus has a score
+    """
+    scores: dict
+
+    def __init__(self, bus):
+        self.scores = {}
+        for word in WORDS:
+            self.scores[word] = 0.0
+        super().__init__(bus)
+
+
+class WEISurveyInformation2023(WEISurveyInformation):
+    """
+    We store the id of the selected bus. We store only the name, but is not used in the selection:
+    that's only for humans that try to read data.
+    """
+    # Random seed that is stored at the first time to ensure that words are generated only once
+    seed = 0
+    step = 0
+
+    def __init__(self, registration):
+        for i in range(1, 21):
+            setattr(self, "word" + str(i), None)
+        super().__init__(registration)
+
+
+class WEISurvey2023(WEISurvey):
+    """
+    Survey for the year 2023.
+    """
+
+    @classmethod
+    def get_year(cls):
+        return 2023
+
+    @classmethod
+    def get_survey_information_class(cls):
+        return WEISurveyInformation2023
+
+    def get_form_class(self):
+        return WEISurveyForm2023
+
+    def update_form(self, form):
+        """
+        Filter the bus selector with the buses of the WEI.
+        """
+        form.set_registration(self.registration)
+
+    @transaction.atomic
+    def form_valid(self, form):
+        word = form.cleaned_data["word"]
+        self.information.step += 1
+        setattr(self.information, "word" + str(self.information.step), word)
+        self.save()
+
+    @classmethod
+    def get_algorithm_class(cls):
+        return WEISurveyAlgorithm2023
+
+    def is_complete(self) -> bool:
+        """
+        The survey is complete once the bus is chosen.
+        """
+        return self.information.step == 20
+
+    @classmethod
+    @lru_cache()
+    def word_mean(cls, word):
+        """
+        Calculate the mid-score given by all buses.
+        """
+        buses = cls.get_algorithm_class().get_buses()
+        return sum([cls.get_algorithm_class().get_bus_information(bus).scores[word] for bus in buses]) / buses.count()
+
+    @lru_cache()
+    def score(self, bus):
+        if not self.is_complete():
+            raise ValueError("Survey is not ended, can't calculate score")
+
+        bus_info = self.get_algorithm_class().get_bus_information(bus)
+        # Score is the given score by the bus subtracted to the mid-score of the buses.
+        s = sum(bus_info.scores[getattr(self.information, 'word' + str(i))]
+                - self.word_mean(getattr(self.information, 'word' + str(i))) for i in range(1, 21)) / 20
+        return s
+
+    @lru_cache()
+    def scores_per_bus(self):
+        return {bus: self.score(bus) for bus in self.get_algorithm_class().get_buses()}
+
+    @lru_cache()
+    def ordered_buses(self):
+        values = list(self.scores_per_bus().items())
+        values.sort(key=lambda item: -item[1])
+        return values
+
+    @classmethod
+    def clear_cache(cls):
+        cls.word_mean.cache_clear()
+        return super().clear_cache()
+
+
+class WEISurveyAlgorithm2023(WEISurveyAlgorithm):
+    """
+    The algorithm class for the year 2023.
+    We use Gale-Shapley algorithm to attribute 1y students into buses.
+    """
+
+    @classmethod
+    def get_survey_class(cls):
+        return WEISurvey2023
+
+    @classmethod
+    def get_bus_information_class(cls):
+        return WEIBusInformation2023
+
+    def run_algorithm(self, display_tqdm=False):
+        """
+        Gale-Shapley algorithm implementation.
+        We modify it to allow buses to have multiple "weddings".
+        """
+        surveys = list(self.get_survey_class()(r) for r in self.get_registrations())  # All surveys
+        surveys = [s for s in surveys if s.is_complete()]  # Don't consider invalid surveys
+        # Don't manage hardcoded people
+        surveys = [s for s in surveys if not hasattr(s.information, 'hardcoded') or not s.information.hardcoded]
+
+        # Reset previous algorithm run
+        for survey in surveys:
+            survey.free()
+            survey.save()
+
+        non_men = [s for s in surveys if s.registration.gender != 'male']
+        men = [s for s in surveys if s.registration.gender == 'male']
+
+        quotas = {}
+        registrations = self.get_registrations()
+        non_men_total = registrations.filter(~Q(gender='male')).count()
+        for bus in self.get_buses():
+            free_seats = bus.size - WEIMembership.objects.filter(bus=bus, registration__first_year=False).count()
+            # Remove hardcoded people
+            free_seats -= WEIMembership.objects.filter(bus=bus, registration__first_year=True,
+                                                       registration__information_json__icontains="hardcoded").count()
+            quotas[bus] = 4 + int(non_men_total / registrations.count() * free_seats)
+
+        tqdm_obj = None
+        if display_tqdm:
+            from tqdm import tqdm
+            tqdm_obj = tqdm(total=len(non_men), desc="Non-hommes")
+
+        # Repartition for non men people first
+        self.make_repartition(non_men, quotas, tqdm_obj=tqdm_obj)
+
+        quotas = {}
+        for bus in self.get_buses():
+            free_seats = bus.size - WEIMembership.objects.filter(bus=bus, registration__first_year=False).count()
+            free_seats -= sum(1 for s in non_men if s.information.selected_bus_pk == bus.pk)
+            # Remove hardcoded people
+            free_seats -= WEIMembership.objects.filter(bus=bus, registration__first_year=True,
+                                                       registration__information_json__icontains="hardcoded").count()
+            quotas[bus] = free_seats
+
+        if display_tqdm:
+            tqdm_obj.close()
+
+            from tqdm import tqdm
+            tqdm_obj = tqdm(total=len(men), desc="Hommes")
+
+        self.make_repartition(men, quotas, tqdm_obj=tqdm_obj)
+
+        if display_tqdm:
+            tqdm_obj.close()
+
+        # Clear cache information after running algorithm
+        WEISurvey2023.clear_cache()
+
+    def make_repartition(self, surveys, quotas=None, tqdm_obj=None):
+        free_surveys = surveys.copy()  # Remaining surveys
+        while free_surveys:  # Some students are not affected
+            survey = free_surveys[0]
+            buses = survey.ordered_buses()  # Preferences of the student
+            for bus, current_score in buses:
+                if self.get_bus_information(bus).has_free_seats(surveys, quotas):
+                    # Selected bus has free places. Put student in the bus
+                    survey.select_bus(bus)
+                    survey.save()
+                    free_surveys.remove(survey)
+                    break
+                else:
+                    # Current bus has not enough places. Remove the least preferred student from the bus if existing
+                    least_preferred_survey = None
+                    least_score = -1
+                    # Find the least student in the bus that has a lower score than the current student
+                    for survey2 in surveys:
+                        if not survey2.information.valid or survey2.information.get_selected_bus() != bus:
+                            continue
+                        score2 = survey2.score(bus)
+                        if current_score <= score2:  # Ignore better students
+                            continue
+                        if least_preferred_survey is None or score2 < least_score:
+                            least_preferred_survey = survey2
+                            least_score = score2
+
+                    if least_preferred_survey is not None:
+                        # Remove the least student from the bus and put the current student in.
+                        # If it does not exist, choose the next bus.
+                        least_preferred_survey.free()
+                        least_preferred_survey.save()
+                        free_surveys.append(least_preferred_survey)
+                        survey.select_bus(bus)
+                        survey.save()
+                        free_surveys.remove(survey)
+                        break
+            else:
+                raise ValueError(f"User {survey.registration.user} has no free seat")
+
+            if tqdm_obj is not None:
+                tqdm_obj.n = len(surveys) - len(free_surveys)
+                tqdm_obj.refresh()

apps/wei/tests/test_wei_algorithm_2023.py

@@ -0,0 +1,110 @@
+# Copyright (C) 2018-2023 by BDE ENS Paris-Saclay
+# SPDX-License-Identifier: GPL-3.0-or-later
+
+import random
+
+from django.contrib.auth.models import User
+from django.test import TestCase
+
+from ..forms.surveys.wei2023 import WEIBusInformation2023, WEISurvey2023, WORDS, WEISurveyInformation2023
+from ..models import Bus, WEIClub, WEIRegistration
+
+
+class TestWEIAlgorithm(TestCase):
+    """
+    Run some tests to ensure that the WEI algorithm is working well.
+    """
+    fixtures = ('initial',)
+
+    def setUp(self):
+        """
+        Create some test data, with one WEI and 10 buses with random score attributions.
+        """
+        self.wei = WEIClub.objects.create(
+            name="WEI 2023",
+            email="wei2023@example.com",
+            date_start='2023-09-16',
+            date_end='2023-09-18',
+            year=2023,
+        )
+
+        self.buses = []
+        for i in range(10):
+            bus = Bus.objects.create(wei=self.wei, name=f"Bus {i}", size=10)
+            self.buses.append(bus)
+            information = WEIBusInformation2023(bus)
+            for word in WORDS:
+                information.scores[word] = random.randint(0, 101)
+            information.save()
+            bus.save()
+
+    def test_survey_algorithm_small(self):
+        """
+        There are only a few people in each bus, ensure that each person has its best bus
+        """
+        # Add a few users
+        for i in range(10):
+            user = User.objects.create(username=f"user{i}")
+            registration = WEIRegistration.objects.create(
+                user=user,
+                wei=self.wei,
+                first_year=True,
+                birth_date='2000-01-01',
+            )
+            information = WEISurveyInformation2023(registration)
+            for j in range(1, 21):
+                setattr(information, f'word{j}', random.choice(WORDS))
+            information.step = 20
+            information.save(registration)
+            registration.save()
+
+        # Run algorithm
+        WEISurvey2023.get_algorithm_class()().run_algorithm()
+
+        # Ensure that everyone has its first choice
+        for r in WEIRegistration.objects.filter(wei=self.wei).all():
+            survey = WEISurvey2023(r)
+            preferred_bus = survey.ordered_buses()[0][0]
+            chosen_bus = survey.information.get_selected_bus()
+            self.assertEqual(preferred_bus, chosen_bus)
+
+    def test_survey_algorithm_full(self):
+        """
+        Buses are full of first year people, ensure that they are happy
+        """
+        # Add a lot of users
+        for i in range(95):
+            user = User.objects.create(username=f"user{i}")
+            registration = WEIRegistration.objects.create(
+                user=user,
+                wei=self.wei,
+                first_year=True,
+                birth_date='2000-01-01',
+            )
+            information = WEISurveyInformation2023(registration)
+            for j in range(1, 21):
+                setattr(information, f'word{j}', random.choice(WORDS))
+            information.step = 20
+            information.save(registration)
+            registration.save()
+
+        # Run algorithm
+        WEISurvey2023.get_algorithm_class()().run_algorithm()
+
+        penalty = 0
+        # Ensure that everyone seems to be happy
+        # We attribute a penalty for each user that didn't have its first choice
+        # The penalty is the square of the distance between the score of the preferred bus
+        # and the score of the attributed bus
+        # We consider it acceptable if the mean of this distance is lower than 5 %
+        for r in WEIRegistration.objects.filter(wei=self.wei).all():
+            survey = WEISurvey2023(r)
+            chosen_bus = survey.information.get_selected_bus()
+            buses = survey.ordered_buses()
+            score = min(v for bus, v in buses if bus == chosen_bus)
+            max_score = buses[0][1]
+            penalty += (max_score - score) ** 2
+
+            self.assertLessEqual(max_score - score, 25)  # Always less than 25 % of tolerance
+
+        self.assertLessEqual(penalty / 100, 25)  # Tolerance of 5 %