Script to create Figure 14 of fourier experiment works

hitchhiking_rotations/cfgs/cfg_pose_to_fourier.py

@@ -7,7 +7,7 @@ def get_cfg_pose_to_fourier(device, nb, nf):
     cfg = {
         "_target_": "hitchhiking_rotations.utils.Trainer",
         "lr": 0.01,
-        "optimizer": "SGD",
+        "optimizer": "Adam",
         "logger": "${logger}",
         "verbose": "${verbose}",
         "device": device,
@@ -20,7 +20,7 @@ def get_cfg_pose_to_fourier(device, nb, nf):
     return {
         "verbose": False,
         "batch_size": 64,
-        "epochs": 10,
+        "epochs": 400,
         "training_data": {
             "_target_": "hitchhiking_rotations.datasets.PoseToFourierDataset",
             "mode": "train",
@@ -40,7 +40,7 @@ def get_cfg_pose_to_fourier(device, nb, nf):
         "val_data": {
             "_target_": "hitchhiking_rotations.datasets.PoseToFourierDataset",
             "mode": "val",
-            "dataset_size": 800,
+            "dataset_size": 400,
             "device": device,
             "nb": nb,
             "nf": nf,

visu/figure_14.py

@@ -0,0 +1,93 @@
+import os
+import numpy as np
+import seaborn as sns
+import matplotlib.pyplot as plt
+from hitchhiking_rotations import HITCHHIKING_ROOT_DIR
+from pathlib import Path
+import pandas as pd
+from hitchhiking_rotations.utils import RotRep
+
+nb_max = 6
+nb_values = range(1, nb_max+1)
+files = [str(s) for nb in nb_values for s in Path(os.path.join(HITCHHIKING_ROOT_DIR, "results", f"pose_to_fourier_{nb}")).rglob("*result.npy")]
+
+#files = [str(s) for s in Path(os.path.join(HITCHHIKING_ROOT_DIR, "results", "pose_to_fourier_1")).rglob("*result.npy")]
+results = [np.load(file, allow_pickle=True) for file in files]
+
+# trainer_name
+methods_res = {}
+selected_metric = "l2"
+rename_and_filter = True
+
+df_res = {}
+df_res["method"] = []
+df_res["score"] = []
+df_res["basis"] = []
+df_res["func"] = []
+
+for ib, run in enumerate(results):
+    for trainer_name, logging_dict in run.items():
+        if trainer_name.find("test") == -1:
+            continue
+
+        # only use metrics generated for testing
+        metrics_test = logging_dict["test"]
+        k = trainer_name[:-5]
+
+        if k not in methods_res:
+            # create a list for each seed
+            methods_res[k] = []
+
+        v = metrics_test[selected_metric]["sum"] / metrics_test[selected_metric]["count"]
+        df_res["method"].append(k)
+        df_res["score"].append(v)
+
+        # File path example: "/hitchhiking-rotations/results/pose_to_fourier_1/seed_2_result.npy"
+        df_res["basis"].append(int(files[ib].split("/")[-2].split("_")[-1]))
+        df_res["func"].append(int(files[ib].split("/")[-1].split("_")[-2]))
+
+        methods_res[k].append(v)
+
+
+df = pd.DataFrame.from_dict(df_res)
+
+if rename_and_filter:
+    mapping = {
+        "r9": RotRep.SVD,
+        "r6": RotRep.GSO,
+        "quat_c": RotRep.QUAT_C,
+        "quat_rf": str(RotRep.QUAT) + "_rf",
+        "rotvec": RotRep.EXP,
+        "euler": RotRep.EULER,
+    }
+
+    training_metric = "l2"
+    df["method"] = df["method"].replace({k + "_" + training_metric: v for k, v in mapping.items()})
+    df["method"] = pd.Categorical(df["method"], categories=[v for v in mapping.values()], ordered=True)
+
+plt.style.use(os.path.join(HITCHHIKING_ROOT_DIR, "assets", "prettyplots.mplstyle"))
+sns.set_style("whitegrid")
+plt.rcParams.update({"font.size": 11})
+
+plt.figure(figsize=(5.5,1.0))
+g = sns.catplot(data=df, x="basis", y="score",
+                hue="method",
+                kind="box",
+                palette='Greens',
+                flierprops={"markeredgecolor": "grey"},
+                height=7.,
+                aspect=2.0)
+
+sns.move_legend(g, "upper left", bbox_to_anchor=(.11, 0.98), ncol=3, title='Network input') # len(names)
+
+for i in range(nb_max-1):
+    plt.axvline(0.5 + i, color="lightgrey", dashes=(2, 2))
+
+plt.xlabel(f"Error - {selected_metric}")
+plt.ylabel("")
+plt.yscale('log')
+plt.tight_layout()
+
+out_p = os.path.join(HITCHHIKING_ROOT_DIR, "results", "figure_14.pdf")
+plt.savefig(out_p)
+plt.show()