diff --git a/mmp/a3/document.pdf b/mmp/a3/document.pdf
index 050dd4c..f3a9191 100644
Binary files a/mmp/a3/document.pdf and b/mmp/a3/document.pdf differ
diff --git a/mmp/a3/document.tex b/mmp/a3/document.tex
index 095db31..25ee4c5 100644
--- a/mmp/a3/document.tex
+++ b/mmp/a3/document.tex
@@ -74,6 +74,8 @@
   \hline Accuracy         & 78.90    \% \\
   \hline
 \end{tabular}
+\\
+The accuracy metric isn’t ideal because it’s quite intuitive to classify whether an object appears once or more in an image. In this case, the model might not have guessed the correct number of objects (e.g., 3 instead of 2), but the result will still be interpreted as accurate. This also makes it more challenging to backpropagate. A better approach would be to output the actual number of objects (in this case, humans).
 
 %------------------ END OF ASSIGNMENT -----------------------
 
diff --git a/mmp/a3/main.py b/mmp/a3/main.py
index bcdaf9c..52c1156 100644
--- a/mmp/a3/main.py
+++ b/mmp/a3/main.py
@@ -15,11 +15,11 @@ def main():
     train_epochs = 10
     model = MmpNet(num_classes=2).to(device=device)
     dataloader_train = get_dataloader(
-        path_to_data="/home/ubuntu/mmp_wise2526_franksim/.data/mmp-public-3.2",
+        path_to_data=".data/mmp-public-3.2",
         image_size=244, batch_size=32, num_workers=6, is_train=True
     )
     dataloader_eval = get_dataloader(
-        path_to_data="/home/ubuntu/mmp_wise2526_franksim/.data/mmp-public-3.2",
+        path_to_data=".data/mmp-public-3.2",
         image_size=244, batch_size=32, num_workers=6, is_train=False
     )
     criterion, optimizer = get_criterion_optimizer(model=model)