a question about time series classification

[Sample-design-alt]

Thanks for your great works!
But I have some question about TSC task.

1. Can time series classification really be unsupervised? According to your paper, use embedding and then SVM for classification, but isn't SVM also a supervised method?
2. Also, I followed your tutorial and I found that using embedding for classification is not even as good as raw data in some datasets.
   （this is GestureMidAirD2 datasets）
   

my code:

---

`
from momentfm import MOMENTPipeline
from torch.utils.data import DataLoader
from momentfm.data.classification_dataset import ClassificationDataset
import numpy as np
model = MOMENTPipeline.from_pretrained(
"AutonLab/MOMENT-1-large",
model_kwargs={
'task_name': 'embedding',
'n_channels': 1,
'num_class': 5
},

)
model.init()

train_dataset = ClassificationDataset(data_split='train')
test_dataset = ClassificationDataset(data_split='test')
train_x = train_dataset.train_data.squeeze()
train_y = train_dataset.train_labels
test_x = test_dataset.test_data.squeeze()
test_y = test_dataset.test_labels

train_dataloader = DataLoader(train_dataset, batch_size=64, shuffle=False, drop_last=False)
test_dataloader = DataLoader(test_dataset, batch_size=64, shuffle=False, drop_last=False)
from tqdm import tqdm
import torch
def get_embedding(model, dataloader):
embeddings, labels = [], []
with torch.no_grad():
for batch_x, batch_masks, batch_labels in tqdm(dataloader, total=len(dataloader)):
batch_x = batch_x.to("cuda").float()
batch_masks = batch_masks.to("cuda")

        output = model(x_enc=batch_x, input_mask=batch_masks) # [batch_size x d_model (=1024)]
        embedding = output.embeddings
        embeddings.append(embedding.detach().cpu().numpy())
        labels.append(batch_labels)

embeddings, labels = np.concatenate(embeddings), np.concatenate(labels)
return embeddings, labels

model.to("cuda").float()

train_embeddings, train_labels = get_embedding(model, train_dataloader)
test_embeddings, test_labels = get_embedding(model, test_dataloader)

print(train_embeddings.shape, train_labels.shape)
print(test_embeddings.shape, test_labels.shape)

from momentfm.models.statistical_classifiers import fit_svm

clf = fit_svm(features=train_embeddings, y=train_labels)
clf1 = fit_svm(features=train_x, y=train_labels)

y_pred_train = clf1.predict(train_x)
y_pred_test = clf1.predict(test_x)
train_accuracy = clf1.score(tra in_x, train_labels)
test_accuracy = clf1.score(test_x, test_labels)

print(f"Train accuracy origin data: {train_accuracy:.2f}")
print(f"Test accuracy origin data: {test_accuracy:.2f}")

y_pred_train = clf.predict(train_embeddings)
y_pred_test = clf.predict(test_embeddings)
train_accuracy = clf.score(train_embeddings, train_labels)
test_accuracy = clf.score(test_embeddings, test_labels)

print(f"Train accuracy embedding by moment: {train_accuracy:.2f}")
print(f"Test accuracy embedding by moment: {test_accuracy:.2f}")
`