add plot script

first_experiment/evaluation.py

@@ -0,0 +1,61 @@
+import glob
+import json
+
+import pandas as pd
+import seaborn as sns
+import matplotlib.pyplot as plt
+
+
+def load_scores() -> pd.DataFrame:
+    path = "/glusterfs/dfs-gfs-dist/goldejon/initial-starting-point-generation"
+    files = glob.glob(path + "/*/results.json")
+
+    scores = {
+        "Dataset Size": [],
+        "Dataset Name": [],
+        "Initialization Strategy": [],
+        "Accuracy": [],
+    }
+
+    for file in files:
+        with open(file, "r") as f:
+            res = json.load(f)
+
+        infos = file.split("/")[-2].split("_")
+
+        if len(infos) == 4:
+            model, dataset, dataset_size, init_strategy = infos
+            embedding_model = None
+        elif len(infos) == 5:
+            model, dataset, dataset_size, init_strategy, embedding_model = infos
+        else:
+            raise ValueError("Wrong number of infos.")
+
+        scores["Dataset Size"].append(int(dataset_size))
+        scores["Dataset Name"].append(dataset)
+        scores["Initialization Strategy"].append(f"{init_strategy} ({embedding_model})")
+        scores["Accuracy"].append(res["test_accuracy"])
+
+    return pd.DataFrame(data=scores)
+
+if __name__ == "__main__":
+    scores = load_scores()
+    full_finetuning = scores[scores["Dataset Size"] == 0]
+    scores = scores[scores["Dataset Size"] > 0]
+
+    sns.set_theme(style="whitegrid")
+    sns.set_context("paper")
+    sns.set(font_scale=1.5)
+
+    g = sns.FacetGrid(scores, col="Dataset Name", sharey=False, height=4, aspect=1.5)
+    g.map(sns.lineplot, "Dataset Size", "Accuracy", "Initialization Strategy", ci=None, marker="o")
+    g.set(xscale="log")
+
+    for idx, col in enumerate(g.col_names):
+        ft_score = full_finetuning[full_finetuning["Dataset Name"] == col]["Accuracy"].iloc[0]
+        g.axes[0][idx].axhline(ft_score, ls="--", color="black")
+
+    g.add_legend()
+
+    plt.show()
+