implemented placeholders

analysis/user_similarity/chronos2/gen_plot.py

@@ -0,0 +1,49 @@
+import os
+import numpy as np
+import datasets
+from glob import glob
+from fire import Fire
+from chronos2 import Chronos2
+
+SleepEDF_PATH = "/mnt/sting/hjyoon/projects/tsllm_personalization_icl/data/SleepEDF"
+CHRONOS2_EMB_PATH = "."
+
+
+def load_data(data_path):
+    return None
+
+
+# def emb_chronos2(data):
+#     return None
+
+def plot_embs(embs):
+    return None
+
+
+def run(
+    data_path=SleepEDF_PATH,
+    is_chached=False,
+    cached_path=CHRONOS2_EMB_PATH,
+):
+    chronos2 = Chronos2()
+    user_test_paths = glob(os.path.join(data_path, "*", "2")) # session 2
+    user_test_data = load_data(user_test_paths)
+    
+    if not is_chached:
+        embs = []
+        for data in user_test_data:
+            emb = chronos2.emb_chronos2(data)
+            embs.append(emb)
+        embs = np.array(embs)
+        np.save(cached_path, embs)
+    else:
+        embs = np.load(cached_path)
+
+    plot_embs(embs)
+    # save plot to pdf in the same directory
+    # tsne speed issue
+    # tsne visualization issue
+
+
+if __name__ == "__main__":
+    Fire(run)

analysis/user_similarity/chronos2/simple_user_classifier.py

@@ -0,0 +1,26 @@
+import os
+from fire import Fire
+import datasets
+
+SleepEDF_PATH = "/mnt/sting/hjyoon/projects/tsllm_personalization_icl/data/SleepEDF"
+
+def run(
+    data_path=SleepEDF_PATH,
+    user_id=None,
+    session_id=None,
+    num_examples=1,
+    num_workers=32,
+    seed=0,
+):
+    user_train_path = os.path.join(data_path, user_id, "1")
+    user_test_path = os.path.join(data_path, user_id, "2")
+    user_train_dataset = datasets.load_from_disk(user_train_path)
+    user_test_dataset = datasets.load_from_disk(user_test_path)
+    user_train_dataset = user_train_dataset.shuffle(seed=seed)
+    user_test_dataset = user_test_dataset.shuffle(seed=seed)
+    user_train_dataset = user_train_dataset.select(range(num_examples))
+    user_test_dataset = user_test_dataset.select(range(num_examples))
+    pass
+
+if __name__ == "__main__":
+    Fire(run)

preprocess/preprocess_SleepEDF.py

@@ -418,7 +418,7 @@ def preprocess(file_path):
         else:
             pbar.update(1)
         w_features = {}
-        missing_w_features = {}
+        raw_data = {}
         flag = False
         for mod in range(num_modalities):
             mod_features = process_by_mod(column_list[mod], x[i, :, mod])
@@ -427,6 +427,8 @@ def preprocess(file_path):
                 break
             for k, v in mod_features.items():
                 w_features[k] = v
+            if "EEG" in column_list[mod]:
+                raw_data[column_list[mod]] = x[i, :, mod]
         if flag:
             continue
         data.append(
@@ -436,6 +438,7 @@ def preprocess(file_path):
                 idx=idx,
                 label=class_dict[y[i]],
                 features=w_features,
+                data=raw_data,
             )
         )
         idx += 1