fix: incorrect max calculation implementation

Author: Essoz

docs/5-min-tutorial.md

@@ -6,7 +6,7 @@ In this tutorial, you will use TrainCheck to detect & diagnose the real‑world
 
 **Prerequisites**  
 - [A working TrainCheck installation](./installation-guide.md)  
-- `efficientnet_pytorch` (install via `pip3 install efficientnet_pytorch`)  
+- `efficientnet_pytorch` and `torchvision` (install via `pip3 install efficientnet_pytorch torchvision`)  
 - A Linux machine with a CUDA‑enabled GPU  
   - 💡 Tip: If you don’t have a CUDA GPU, you can still run this tutorial on CPU—it’ll just take longer.
 

traincheck/invariant/lead_relation.py

@@ -428,7 +428,7 @@ def generate_hypothesis(trace) -> list[Hypothesis]:
                 assert last_example is not None
 
                 if event_A_idx >= len(events_A_pre):
-                    max_time = events_B_post["time"].max()
+                    max_time = max([event["time"] for event in events_B_post])
                     if pre_event_A_time <= max_time:
                         hypothesis_with_examples[
                             (func_A, func_B)
@@ -673,7 +673,7 @@ def collect_examples(trace, hypothesis):
                     break
 
                 if event_A_idx >= len(events_A_pre):
-                    max_time = events_B_post["time"].max()
+                    max_time = max([event["time"] for event in events_B_post])
                     if pre_event_A_time <= max_time:
                         hypothesis[(func_A, func_B)].positive_examples.add_example(
                             last_example