add script to plot score distributions for several alpha values

backend/scripts/research/alpha_plots.py

@@ -0,0 +1,113 @@
+from os import environ
+
+import pandas as pd
+import numpy as np
+import seaborn as sns
+from matplotlib import pyplot as plt
+from fabric import Connection
+
+PG_HOST = "localhost"
+PG_USER = "tournesol"
+PG_PWD = environ["PG_PWD"]
+PG_DB = "tournesol"
+
+PG_CONNECTION_STRING = f"postgresql://{PG_USER}:{PG_PWD}@{PG_HOST}/{PG_DB}"
+
+
+def indiv_plot(alpha, with_poll_scaling=False):
+    sql_indiv = """
+    WITH comparisons_count AS (
+        SELECT e.id as entity_id, user_id, COUNT(*) as n_comparisons
+            FROM tournesol_entity e 
+            INNER JOIN tournesol_comparison c ON e.id = c.entity_1_id OR e.id = c.entity_2_id
+            GROUP BY e.id, c.user_id
+    )
+    SELECT  cr.entity_id,
+            cr.user_id,
+            score AS poll_score, 
+            raw_score * cs.scale + cs.translation AS scaled_score,
+            cc.n_comparisons 
+        FROM tournesol_contributorratingcriteriascore crcs
+        INNER JOIN tournesol_contributorrating cr on cr.id = crcs.contributor_rating_id
+        INNER JOIN comparisons_count cc ON cc.entity_id = cr.entity_id AND cc.user_id = cr.user_id
+        INNER JOIN tournesol_contributorscaling cs ON cs.poll_id = cr.poll_id AND cs.user_id = cr.user_id AND cs.criteria = crcs.criteria
+        WHERE crcs.criteria = 'largely_recommended'
+    """
+    df_indiv = pd.read_sql(sql_indiv, con=PG_CONNECTION_STRING)
+    df_indiv["n_comparisons"] = pd.cut(df_indiv.n_comparisons, bins=[0,1,2,4,8,np.inf])
+    plt.figure(figsize=(15,4.5))
+    ax = sns.histplot(
+        data=df_indiv,
+        x="poll_score" if with_poll_scaling else "scaled_score",
+        hue="n_comparisons",
+        hue_order=sorted(df_indiv["n_comparisons"].unique(), reverse=True),
+        palette="coolwarm_r",
+        multiple="stack",
+        **(
+            {
+                "binwidth": 2,
+                "binrange": [-100,100]
+            } 
+            if with_poll_scaling
+            else {
+                "binwidth": 0.01,
+                "binrange": [-0.9, 0.9]
+            }
+        )
+    )
+
+    if with_poll_scaling:
+        ax.set_title(f"Scaled individual scores (after poll scaling) | alpha = {alpha:.2f}")
+    else:
+        ax.set_title(f"Scaled individual scores | alpha = {alpha:.2f}")
+
+    plt.savefig(f"indiv_scores_{ 'rescaled_' if with_poll_scaling else '' }{alpha:.2f}.png", dpi=150)
+
+    return ax
+
+
+def global_plot(alpha):
+    sql =  """
+    SELECT uid, tournesol_score, rating_n_contributors
+        FROM tournesol_entity e
+        WHERE tournesol_score is not null and type = 'video'
+    """
+
+    df = pd.read_sql(sql, con=PG_CONNECTION_STRING)
+    df["n_contributors"] = df.rating_n_contributors.map(lambda x: str(x) if x <=3 else "4+")
+    plt.figure(figsize=(15,5))
+    ax = sns.histplot(
+        data=df,
+        x="tournesol_score",
+        hue="n_contributors",
+        hue_order=sorted(df["n_contributors"].unique()),
+        palette="coolwarm",
+        multiple="stack",
+        binwidth=2,
+        binrange=[-100,100],
+    )
+    ax.set_title(f"Global scores | alpha = {alpha:.2f}")
+
+    plt.xlabel("score")
+    plt.savefig(f"global_scores_{alpha:.2f}.png", dpi=150)
+
+    return ax
+
+
+def run_ml(alpha):
+    alpha = float(alpha)
+    conn = Connection("tournesol-vm")
+    with conn.cd("/srv/tournesol-backend"):
+        conn.run(
+            "sudo -u gunicorn SETTINGS_FILE=/etc/tournesol/settings.yaml "
+            "./venv/bin/python manage.py ml_train --main-criterion-only "
+            f" --alpha={alpha}"
+        )
+
+
+def all_plots():
+    for alpha in [0.1, 1.0, 10.0]:
+        run_ml(alpha)
+        global_plot(alpha)
+        indiv_plot(alpha, with_poll_scaling=True)
+        indiv_plot(alpha, with_poll_scaling=False)