File size: 7,031 Bytes
dbd3b18
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
import json
import os
import re

def safe_float(value):
    """Convert a value to float safely. Returns None if conversion fails."""
    try:
        return float(value)
    except ValueError:
        return None


def calculate_task_metrics(task_info):
    """Calculate average accuracy, best prompt, and CPS for a task."""
    accuracies = [prompt['value'] for prompt in task_info['prompts'] if prompt['value'] is not None]

    if not accuracies:
        return None

    task_info['average_accuracy'] = sum(accuracies) / len(accuracies)
    best_prompt_data = max(task_info['prompts'], key=lambda x: x['value'])
    task_info['best_prompt'] = best_prompt_data['value']
    task_info['prompt_id'] = best_prompt_data['prompt']

    # Calculate CPS
    avg_acc = task_info['average_accuracy']
    best_acc = task_info['best_prompt']
    task_info['CPS'] = (1 - (best_acc - avg_acc) / 100) * best_acc


def extract_data_from_file(file_path):
    """Extract task and prompt data from the given file."""
    with open(file_path, 'r') as file:
        lines = file.readlines()

    tasks_data = {}
    current_task = None

    for line in lines:
        line = line.strip()

        # Skip irrelevant lines
        if not line:
            continue


        if line.startswith("|         Tasks"):
            continue

        if line.startswith("hf (pretrained="):

            # Estrai la parte dopo "pretrained="
            start = line.find("pretrained=") + len("pretrained=")
            end = line.find(",", start)  # Trova la virgola successiva
            # Estrai la stringa desiderata
            pretrained_model = line[start:end]

            # Estrarre num_fewshot
            num_fewshot_match = re.search(r"num_fewshot:\s*([\w\d]+)", line)
            num_fewshot = num_fewshot_match.group(1) if num_fewshot_match else None

            # Estrarre batch_size
            batch_size_match = re.search(r"batch_size:\s*(\d+)", line)
            batch_size = int(batch_size_match.group(1)) if batch_size_match else None

            continue

        columns = line.split('|')
        if len(columns) != 11:
            continue

        task_name = columns[1]
        metric = columns[5].strip()
        value = safe_float(columns[7])
        stderr = safe_float(columns[9])

        if metric == "acc_norm":
            continue

        # Identify task and prompts
        if task_name.startswith(" - "):
            task_name = task_name[3:].strip()
            current_task = task_name
            tasks_data.setdefault(current_task,
                                  {'prompts': [], 'average_accuracy': 0, 'best_prompt': None, 'prompt_id': None,
                                   'CPS': None})

        elif task_name.startswith("  - ") and current_task:
            prompt_name = task_name[4:].strip()
            prompt_data = {'prompt': prompt_name, 'metric': metric, 'value': value * 100,
                           'stderr': stderr}
            tasks_data[current_task]['prompts'].append(prompt_data)

    # Special handling for evalita NER
    if "evalita NER" in tasks_data:
        task_info = tasks_data["evalita NER"]
        weight_map = {"ADG prompt-1": 521, "ADG prompt-2": 521, "FIC prompt-1": 1517, "FIC prompt-2": 1517,
                      "WN prompt-1": 2088, "WN prompt-2": 2088}

        weighted_values = {"prompt-1": 0, "prompt-2": 0}
        total_weights = sum(weight_map.values())

        for prompt in task_info['prompts']:
            if prompt['prompt'] in weight_map:
                if "prompt-1" in prompt['prompt']:
                    weighted_values["prompt-1"] += weight_map[prompt['prompt']] * prompt['value']
                elif "prompt-2" in prompt['prompt']:
                    weighted_values["prompt-2"] += weight_map[prompt['prompt']] * prompt['value']

        task_info['prompts'] = [
            {"prompt": "prompt-1", "metric": "acc", "value": weighted_values["prompt-1"] / total_weights,
             'stderr': None},
            {"prompt": "prompt-2", "metric": "acc", "value": weighted_values["prompt-2"] / total_weights,
             'stderr': None}]

    # Calculate metrics for each task
    for task_info in tasks_data.values():
        calculate_task_metrics(task_info)

    # Calculate average CPS
    tasks_with_cps = [task['CPS'] for task in tasks_data.values() if task['CPS'] is not None]
    average_CPS = sum(tasks_with_cps) / len(tasks_with_cps) if tasks_with_cps else 0

    config = {
        "model_name": pretrained_model,
        "num_fewshot": num_fewshot,
        "batch_size": batch_size
    }

    return {'average_CPS': average_CPS, 'config': config, 'tasks': tasks_data}


# Example usage
#file_path = '../evalita_llm_results/models_output/slurm-7769.out'
#json_output = extract_data_from_file(file_path)
#print(json_output)


# Directory da cui leggere i file .out
directory_in_path = '../evalita_llm_models_output/'
directory_out_results_path = '../evalita_llm_results/'
directory_out_requests_path = '../evalita_llm_requests/'

# Itera sui file nella directory
for filename in os.listdir(directory_in_path):
    if filename.endswith('.out'):
        # Costruisci il percorso completo del file
        file_path = os.path.join(directory_in_path, filename)

        # Esegui la funzione extract_data_from_file
        json_output = extract_data_from_file(file_path)

        # Estrai model_org_name e model_name da model_name
        model_org_name, model_name = json_output['config']['model_name'].split('/')






        # Percorso del file JSON di configurazione in ../evalita_llm_requests2/
        config_file_path = os.path.join(directory_out_requests_path, model_org_name, f"{model_name}.json")

        # Se il file esiste, caricalo e aggiorna il dizionario config
        if os.path.exists(config_file_path):
            with open(config_file_path, 'r', encoding='utf-8') as config_file:
                additional_config = json.load(config_file)

            # Aggiorna la configurazione con i nuovi dati
            json_output['config'].update(additional_config)




        # Crea il percorso della cartella per model_org_name
        org_folder_path = os.path.join(directory_out_results_path, model_org_name)
        os.makedirs(org_folder_path, exist_ok=True)  # Crea la cartella se non esiste

        # Crea il percorso completo del file JSON
        file_suffix = f"{json_output['config']['num_fewshot']}"
        output_file_path = os.path.join(org_folder_path, f"{model_name}_{file_suffix}.json")

        # Salva il JSON in un file con ritorni a capo compatibili con Linux
        with open(output_file_path, 'w', newline="\n") as outfile:
            json.dump(json_output, outfile, indent=4)

        # Stampa il risultato
        print(f"File {filename} elaborato e salvato in {output_file_path}")