dataset estimates

src/data/dataset.py

@@ -36,8 +36,8 @@ def __init__(self, path: str, chunk_num: int, train: bool, device: str, dataset_
         # used to reduce dataset size, for better training on low end devices
         new_dataset_len = int(
             len(self.frame_paths)*dataset_percentage/100.0)
-        print(
-            f"using {new_dataset_len}/{len(self.frame_paths)} frames for training")
+        # print(
+        #     f"using {new_dataset_len}/{len(self.frame_paths)} frames for training")
         self.frame_paths = self.frame_paths[:new_dataset_len]
 
         # the parent dir of the parent dir

src/dataset_params.py

@@ -0,0 +1,40 @@
+import torch
+from torch.utils.data import DataLoader
+from tabulate import tabulate
+
+from data.dataset import CommaDataset
+from config import cfg
+
+# params
+batch_size = 4
+dataset_percentage = 10
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+# data
+train_dataset = CommaDataset(
+    cfg["data"]["path"], chunk_num=1, train=True, device=device, dataset_percentage=dataset_percentage
+)
+train_dataloader = DataLoader(
+    train_dataset, batch_size=batch_size, shuffle=True, pin_memory=True, num_workers=4)
+
+
+def iters_to_time(iters: int, per_batch: float = 0.177) -> float:
+    # per_batch = 0.089 # pilotnet
+    sec_per_iter = per_batch*batch_size
+    sec = iters * sec_per_iter
+    hours = sec / 3600.0
+    return hours
+
+
+one_epoch = len(train_dataloader)
+epochs_num = 20
+total_iters = one_epoch * epochs_num
+
+# print results
+print(tabulate([
+    [f"{dataset_percentage}% of dataset", len(train_dataloader)],
+    ["batch size", batch_size],
+    [f"total epochs", epochs_num],
+    ["time (hours)", iters_to_time(total_iters)],
+], floatfmt=".2f"))