add assess.py, pass the json from run.py here

assess.py

@@ -0,0 +1,94 @@
+from datetime import datetime, timedelta
+import json
+import math
+def analyze_time_difference(ground_truth: str, to_assess: str) -> dict:
+    def time_to_minutes(time_str: str) -> int:
+        """Convert HH:MM to total minutes since midnight"""
+        hours, minutes = map(int, time_str.split(':'))
+        return hours * 60 + minutes
+    
+    def minutes_to_time(minutes: int) -> str:
+        """Convert total minutes to HH:MM format"""
+        hours = (minutes // 60) % 24
+        mins = minutes % 60
+        return f"{hours:02d}:{mins:02d}"
+    
+    def get_minute_difference(time1: str, time2: str) -> int:
+        """Calculate absolute minute difference between two times"""
+        minutes1 = time_to_minutes(time1)
+        minutes2 = time_to_minutes(time2)
+        
+        # Handle cases crossing midnight
+        diff = abs(minutes1 - minutes2)
+        return min(diff, 1440 - diff)  # 1440 = 24*60
+    
+    # Calculate normal difference
+    normal_diff = get_minute_difference(ground_truth, to_assess)
+    
+    # Switch hour and minute hands for to_assess
+    hours, minutes = map(int, to_assess.split(':'))
+    if minutes >= 12:  # Convert minute hand position to valid hour
+        minutes = minutes % 12
+    switched_time = f"{minutes:02d}:{hours:02d}"
+    
+    # Calculate difference with switched hands
+    switched_diff = get_minute_difference(ground_truth, switched_time)
+    
+    return {
+        "normal_diff_minutes": normal_diff,
+        "switched_diff_minutes": switched_diff
+    }
+
+
+
+
+
+with open('molmo-answers-72b.json', 'r') as file:
+    input_data = json.load(file)
+    
+
+with open('benchmark.json', 'r') as file:
+    ground_truth = json.load(file)
+
+
+
+in_5_mins = 0
+in_4_mins = 0
+in_3_mins = 0
+in_2_mins = 0
+in_1_min = 0
+exact = 0
+
+i = 0
+
+while i < len(ground_truth):
+    human_answer = ground_truth[i]['human_time']
+    smallest = 10000000
+    to_use = {'normal_diff_minutes': 10000000}
+    for t in input_data[i]['reported_times']:
+        assessment = analyze_time_difference(human_answer, t)
+        to_use['normal_diff_minutes'] = min(to_use['normal_diff_minutes'], assessment['normal_diff_minutes'])
+
+    assessment = to_use
+    if assessment['normal_diff_minutes'] == 0:
+        exact += 1
+    if assessment['normal_diff_minutes'] <= 1:
+        in_1_min += 1
+    if assessment['normal_diff_minutes'] <= 2:
+        in_2_mins += 1
+    if assessment['normal_diff_minutes'] <= 3:
+        in_3_mins += 1
+    if assessment['normal_diff_minutes'] <= 4:
+        in_4_mins += 1
+    if assessment['normal_diff_minutes'] <= 5:
+        in_5_mins += 1
+
+    i += 1
+
+print('Molmo 72b benchmark')
+print('Within 5 mins ->', f'{in_5_mins/len(ground_truth):.2f}')
+print('Within 4 mins ->', f'{in_4_mins/len(ground_truth):.2f}')
+print('Within 3 mins ->', f'{in_3_mins/len(ground_truth):.2f}')
+print('Within 2 mins ->', f'{in_2_mins/len(ground_truth):.2f}')
+print('Within 1 min ->', f'{in_1_min/len(ground_truth):.2f}')
+print('Exact ->', f'{exact/len(ground_truth):.2f}')