feat: add plot rmse

src/eval/logger.py

@@ -1,12 +1,15 @@
 from datetime import datetime
+from pathlib import Path
 from typing import Literal
-
+import pandas
 import torch
 from matplotlib import pyplot as plt
+from pandas import DataFrame
 from torch import Tensor
 
 import wandb
 
+from .utils import calculate_RMSE
 from ..my_gsplat.geometry import compute_silhouette_diff
 
 
@@ -19,13 +22,16 @@ def __init__(self, run_name: str | None = None, config: dict = None):
         if run_name is None:
             run_name = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
         else:
-            run_name = run_name + "_" + datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
+            run_name = run_name
+        self.entity = "supavision"
+        self.project = "ABGICP"
         wandb.init(
-            project="ABGICP",
-            entity="supavision",
+            project=self.project,
+            entity=self.entity,
             name=run_name,
             config=config,
         )
+        self.api = wandb.Api()
 
         print(f"Run name: {run_name}:config: {config}")
 
@@ -227,3 +233,69 @@ def plot_rgbd(
         wandb.log({fig_title: wandb.Image(fig)}, step=step)
 
         plt.close()
+
+    def plot_bar(
+        self,
+        plot_name: str,
+        label_name: str,
+        value_name: str,
+        labels: list,
+        values: list,
+    ):
+
+        data = [[label, val] for (label, val) in zip(labels, values)]
+        table = wandb.Table(data=data, columns=[label_name, value_name])
+        wandb.log(
+            {plot_name: wandb.plot.bar(table, label_name, value_name, title=plot_name)}
+        )
+
+    def load_history(self, tags: str = "gsplatloc") -> dict[DataFrame]:
+        """
+        https://docs.wandb.ai/ref/python/public-api/api#runs
+        Parameters
+        ----------
+        tags: str
+
+        Returns
+        -------
+        histories: dict[str(sub_set),run.history]
+        """
+        # filter tags runs
+        _runs = self.api.runs(
+            path="supavision/ABGICP", filters={"tags": tags}, order="config.+.sub_set"
+        )
+        histories = {}
+        for _run in _runs:
+            run_path = Path(*_run.path).as_posix()
+            run = self.api.run(path=run_path)
+            histories[run.config["sub_set"]] = run.history(2000)
+        assert len(histories) == len(_runs)
+        return histories
+
+    def plot_RMSE(self):
+        histories = self.load_history()
+        ates = []
+        ares = []
+        scenes = []
+        # Read and calculate RMSEs
+        for scene, his in histories.items():
+            scenes.append(scene)
+            eT = his["Translation Error"].to_numpy()
+            eR = his["Rotation Error"].to_numpy()
+            ates.append(calculate_RMSE(eT))
+            ares.append(calculate_RMSE(eR))
+
+        self.plot_bar(
+            "ATE of Replica",
+            label_name="scenes",
+            value_name="ATEs",
+            values=ates,
+            labels=scenes,
+        )
+        self.plot_bar(
+            "ARE of Replica",
+            label_name="scenes",
+            value_name="AREs",
+            values=ares,
+            labels=scenes,
+        )

src/eval/utils.py

@@ -1,5 +1,5 @@
 import numpy as np
-from numpy._typing import NDArray
+from numpy.typing import NDArray
 
 
 def calculate_translation_error(
@@ -99,3 +99,12 @@ def diff_pcd_COM(pcd_1: NDArray[np.float64], pcd_2: NDArray[np.float64]) -> floa
     com2 = np.mean(pcd_2, axis=0)
     distance = np.linalg.norm(com1 - com2)
     return distance
+
+
+def calculate_RMSE(eT: NDArray) -> float:
+    """
+    Returns
+    -------
+    RMSE: float
+    """
+    return np.sqrt(np.mean(np.square(eT)))

src/plot_rmse.py

@@ -0,0 +1,6 @@
+from src.eval.logger import WandbLogger
+
+if __name__ == "__main__":
+    log = WandbLogger(run_name="plot_RMSE", config={"description": "plots"})
+
+    log.plot_RMSE()