From 841d54c050269d3d548c418feeb1eafaa34600fc Mon Sep 17 00:00:00 2001
From: Pepijn <pepijn@huggingface.co>
Date: Tue, 11 Mar 2025 12:23:51 +0100
Subject: [PATCH] Use sampler always (temp fix)

---
 lerobot/scripts/train.py | 24 ++++++++++++++++--------
 1 file changed, 16 insertions(+), 8 deletions(-)

diff --git a/lerobot/scripts/train.py b/lerobot/scripts/train.py
index 52fed33de..1ad16a153 100644
--- a/lerobot/scripts/train.py
+++ b/lerobot/scripts/train.py
@@ -25,7 +25,7 @@ from torch.amp import GradScaler
 from torch.optim import Optimizer
 
 from lerobot.common.datasets.factory import make_dataset
-from lerobot.common.datasets.sampler import PrioritizedSampler
+from lerobot.common.datasets.sampler import EpisodeAwareSampler, PrioritizedSampler
 from lerobot.common.datasets.utils import cycle
 from lerobot.common.envs.factory import make_env
 from lerobot.common.optim.factory import make_optimizer_and_scheduler
@@ -166,18 +166,26 @@ def train(cfg: TrainPipelineConfig):
     # create dataloader for offline training
     if hasattr(cfg.policy, "drop_n_last_frames"):
         shuffle = False
-        sampler = PrioritizedSampler(
-            data_len=data_len,
-            alpha=0.6,
-            beta=0.1,
-            eps=1e-6,
-            replacement=True,
-            num_samples_per_epoch=data_len,
+        sampler = EpisodeAwareSampler(
+            dataset.episode_data_index,
+            drop_n_last_frames=cfg.policy.drop_n_last_frames,
+            shuffle=True,
         )
     else:
         shuffle = True
         sampler = None
 
+    # TODO(pepijn): If experiment works integrate this
+    shuffle = False
+    sampler = PrioritizedSampler(
+        data_len=data_len,
+        alpha=0.6,
+        beta=0.1,
+        eps=1e-6,
+        replacement=True,
+        num_samples_per_epoch=data_len,
+    )
+
     dataloader = torch.utils.data.DataLoader(
         dataset,
         num_workers=cfg.num_workers,