Initial commit

Dockerfile

@@ -0,0 +1,9 @@
+FROM python:3.9
+
+COPY requirements.txt .
+
+RUN pip install --no-cache-dir --upgrade -r requirements.txt
+
+COPY . .
+
+CMD ["python main.py"]

main.py

@@ -0,0 +1,137 @@
+import random
+import pandas as pd
+from datetime import datetime
+
+
+class Model:
+    """
+    Class containing the info of a model.
+
+    :param name: Name of the model
+    :param elo: Elo rating of the model
+    :param games_played: Number of games played by the model (useful if we implement sigma uncertainty)
+    """
+    def __init__(self, name, elo=1200, games_played=0):
+        self.name = name
+        self.elo = elo
+        self.games_played = games_played
+
+
+class Matchmaking:
+    """
+    Class managing the matchmaking between the models.
+
+    :param models: List of models
+    :param queue: Temporary list of models used for the matching process
+    :param k: Dev coefficient
+    :param max_diff: Maximum difference considered between two models' elo
+    :param matches: Dictionary containing the match history (to later upload as CSV)
+    """
+    def __init__(self, models):
+        self.models = models
+        self.queue = self.models.copy()
+        self.k = 20
+        self.max_diff = 500
+        self.matches = {
+            "model1": [],
+            "model2": [],
+            "result": [],
+            "datetime": []
+        }
+
+    def run(self):
+        """
+        Run the matchmaking process.
+        Add models to the queue, shuffle it, and match the models one by one to models with close ratings.
+        Compute the new elo for each model after each match and add the match to the match history.
+        """
+        for i in range(10):
+            self.queue = self.models.copy()
+            random.shuffle(self.queue)
+            while len(self.queue) > 1:
+                model1 = self.queue.pop(0)
+                model2 = self.queue.pop(self.find_n_closest_indexes(model1, 10))
+                result = match(model1, model2)
+                self.compute_elo(model1, model2, result)
+                self.matches["model1"].append(model1.name)
+                self.matches["model2"].append(model2.name)
+                self.matches["result"].append(result)
+                self.matches["datetime"].append(datetime.now().strftime("%Y-%m-%d %H:%M:%S.%f"))
+
+    def compute_elo(self, model1, model2, result):
+        """ Compute the new elo for each model based on a match result. """
+        delta = model1.elo - model2.elo
+        win_probability = 1 / (1 + 10 ** (-delta / 500))
+        model1.elo += self.k * (result - win_probability)
+        model2.elo -= self.k * (result - win_probability)
+
+    def find_n_closest_indexes(self, model, n) -> int:
+        """
+        Get a model index with a fairly close rating. If no model is found, return the last model in the queue.
+        We don't always pick the closest rating to add variety to the matchups.
+
+        :param model: Model to compare
+        :param n: Number of close models from which to pick a candidate
+        :return: id of the chosen candidate
+        """
+        indexes = []
+        closest_diffs = [9999999] * n
+        for i, m in enumerate(self.queue):
+            if m.name == model.name:
+                continue
+            diff = abs(m.elo - model.elo)
+            if diff < max(closest_diffs):
+                closest_diffs.append(diff)
+                closest_diffs.sort()
+                closest_diffs.pop()
+                indexes.append(i)
+        random.shuffle(indexes)
+        return indexes[0]
+
+    def to_csv(self):
+        """ Save the match history as a CSV file to the hub. """
+        df = pd.DataFrame(columns=['name', 'elo'])
+        for model in self.models:
+            df = pd.concat([df, pd.DataFrame([[model.name, model.elo]], columns=['name', 'elo'])])
+        df.to_csv('elo.csv', index=False)
+        df_matches = pd.DataFrame(self.matches)
+        df_matches.to_csv(f"matches/{datetime.now().strftime('%Y-%m-%d_%H-%M-%S_%f')}.csv", index=False)
+
+
+def match(model1, model2) -> float:
+    """
+    !!! Current code is placeholder !!!
+    TODO: Launch a Unity process with the 2 models and get the result of the match
+
+    :param model1: First Model object
+    :param model2: Second Model object
+    :return: match result (0: model1 lost, 0.5: draw, 1: model1 won)
+    """
+    result = random.randint(0, 2) / 2
+    return result
+
+
+def get_models_list() -> list:
+    """
+    !!! Current code is placeholder !!!
+    TODO: Create a list of Model objects from the models found on the hub
+
+    :return: list of Model objects
+    """
+    models = []
+    data = pd.read_csv("example.csv")
+    for i, row in data.iterrows():
+        models.append(Model(row["name"], row["elo"]))
+    return models
+
+
+def init_matchmaking():
+    models = get_models_list()
+    matchmaking = Matchmaking(models)
+    matchmaking.run()
+    matchmaking.to_csv()
+
+
+if __name__ == "__main__":
+    print("It's running!")
+    # init_matchmaking()