--- a
+++ b/performance.py
@@ -0,0 +1,96 @@
+import os
+from pathlib import Path
+
+import pandas as pd
+import torch
+
+from metrics import get_all_metrics, get_regression_metrics
+
+def export_performance(
+    src_path: str,
+    dst_root: str='performance',
+):
+    logits = pd.read_pickle(src_path)
+    config = logits['config']
+    if config['task'] == 'multitask':
+        _labels = logits['labels']
+        _preds = logits['preds']
+        labels = []
+        preds = []
+        for label, pred in zip(_labels, _preds):
+            if pred[0] != 0.501:
+                labels.append(label)
+                preds.append(pred)
+        _labels, _preds, labels, preds = torch.tensor(_labels), torch.tensor(_preds), torch.tensor(labels), torch.tensor(preds)
+        outcome_metrics = get_all_metrics(preds[:, 0], labels[:, 0], 'outcome', None)
+        readmission_metrics = get_all_metrics(preds[:, 1], labels[:, 1], 'outcome', None)
+        _outcome_metrics = get_all_metrics(_preds[:, 0], _labels[:, 0], 'outcome', None)
+        _readmission_metrics = get_all_metrics(_preds[:, 1], _labels[:, 1], 'outcome', None)
+        data = {'count': [len(_labels), len(labels)] * 2}
+        data = dict(data, **{k: [v1, v2, v3, v4] for k, v1, v2, v3, v4 in zip(_outcome_metrics.keys(), _outcome_metrics.values(), outcome_metrics.values(), _readmission_metrics.values(), readmission_metrics.values())})
+        performance = pd.DataFrame(data=data, index=['o all', 'o without unknown samples', 'r all', 'r without unknown samples'])
+    elif config['task'] == 'los':
+        _labels = logits['labels']
+        _preds = logits['preds']
+        labels = []
+        preds = []
+        for label, pred in zip(_labels, _preds):
+            if pred[0] != 0:
+                labels.append(label)
+                preds.append(pred)
+        data = {'count': [len(_labels), len(labels)]}
+        _labels = torch.vstack([torch.tensor(label).unsqueeze(1) for label in _labels]).squeeze(-1)
+        _preds = torch.vstack([torch.tensor(pred).unsqueeze(1) for pred in _preds]).squeeze(-1)
+        labels = torch.vstack([torch.tensor(label).unsqueeze(1) for label in labels]).squeeze(-1)
+        preds = torch.vstack([torch.tensor(pred).unsqueeze(1) for pred in preds]).squeeze(-1)
+        _metrics = get_regression_metrics(_preds, _labels)
+        metrics = get_regression_metrics(preds, labels)
+        data = dict(data, **{k: [f'{v1:.2f}', f'{v2:.2f}'] for k, v1, v2 in zip(_metrics.keys(), _metrics.values(), metrics.values())})
+        performance = pd.DataFrame(data=data, index=['all', 'w/o'])
+    else:
+        _labels = logits['labels']
+        _preds = logits['preds']
+        _metrics = get_all_metrics(_preds, _labels, 'outcome', None)
+        labels = []
+        preds = []
+        for label, pred in zip(_labels, _preds):
+            if pred != 0.501:
+                labels.append(label)
+                preds.append(pred)
+        metrics = get_all_metrics(preds, labels, 'outcome', None)
+        data = {'count': [len(_labels), len(labels)]}
+        data = dict(data, **{k: [f'{v1 * 100:.2f}', f'{v2 * 100:.2f}'] for k, v1, v2 in zip(_metrics.keys(), _metrics.values(), metrics.values())})
+    
+        performance = pd.DataFrame(data=data, index=['all', 'without unknown samples'])
+    
+    time = config.get('time', 0)
+    if time == 0:
+        time_des = 'upon-discharge'
+    elif time == 1:
+        time_des = '1month'
+    elif time == 2:
+        time_des = '6months'
+    dst_path = os.path.join(dst_root, config['dataset'], config['task'], config['model'])
+    sub_dst_name = f'{config["form"]}_{str(config["n_shot"])}shot_{time_des}'
+    if config['unit'] is True:
+        sub_dst_name += '_unit'
+    if config['reference_range'] is True:
+        sub_dst_name += '_range'
+    if config.get('prompt_engineering') is True:
+        sub_dst_name += '_cot'
+    impute = config.get('impute')
+    if impute == 0:
+        sub_dst_name += '_no_impute'
+    elif impute == 1:
+        sub_dst_name += '_impute'
+    elif impute == 2:
+        sub_dst_name += '_impute_info'
+    Path(dst_path).mkdir(parents=True, exist_ok=True)
+    performance.to_csv(os.path.join(dst_path, f'{sub_dst_name}.csv'))
+
+if __name__ == '__main__':
+    for file in [
+        'logits/tjh/outcome/gpt-3.5-turbo-1106/string_1shot_upon-discharge_unit_range_impute.pkl',
+        'logits/tjh/outcome/gpt-3.5-turbo-1106/string_1shot_upon-discharge_unit_range_no_impute.pkl'
+    ]:
+        export_performance(file)
\ No newline at end of file