Update README (#3)

- Update README (12144fd7ab4ed55436aa315f444729e47d1ffa6a)

README.md

@@ -1,11 +1,11 @@
 ---
-task_categories:
-- text-generation
 license: cc-by-sa-4.0
 language:
 - en
 tags:
 - music
+size_categories:
+- 1K<n<10K
 ---
 # Google/Music-Capsの音声データをスペクトログラム化したデータ。
 
@@ -26,13 +26,14 @@ data = data["train"]
 ```
 
 ### 1: データローダーへ
-* まだテストデータと検証データは用意していないので、コメントアウトしています
 * こんな感じの関数で、データローダーにできます。
 ```py
 from torchvision import transforms
 from torch.utils.data import DataLoader
 BATCH_SIZE = ??? # 自分で設定
 IMAGE_SIZE = ???
+TRAIN_SIZE = ??? # 訓練に使用したいデータセット数
+TEST_SIZE = ??? # テストに使用したいデータセット数
 
 def load_datasets():
     data_transforms = [
@@ -42,26 +43,24 @@ def load_datasets():
     ]
     data_transform = transforms.Compose(data_transforms)
     
-    train = load_dataset("mickylan2367/spectrogram", split="train")
-    # test = load_dataset("mickylan2367/spectrogram", split="test")
-    # validation = load_dataset("mickylan2367/spectrogram", split="validation")
+    data = load_dataset("mickylan2367/spectrogram")
+    data = data["train"]
+    train = data[slice(0, TRAIN_SIZE, None)]
+    test = data[slice(TRAIN_SIZE, TRAIN_SIZE + TEST_SIZE, 0)]
     
     for idx in range(len(train["image"])):
         train["image"][idx] = data_transform(train["image"][idx])
-        # test["image"][idx] = data_transform(test["image"][idx])
-        # validation["image"][idx] = data_transform(validation["image"][idx])
+        test["image"][idx] = data_transform(test["image"][idx])
     
     train = Dataset.from_dict(train)
-    # test = Dataset.from_dict(test)
-    # validation = Dataset.from_dict(validation)
-    
     train = train.with_format("torch") # リスト型回避
-    # test = test.with_format("torch")
-    # validation = validation.with_format(validation)
-    
-    # dataset = torch.utils.data.ConcatDataset([train, validation, test])
-    # dataloader = DataLoader(dataset, batch_size=BATCH_SIZE, shuffle=True, drop_last=True)
-    # return dataloader
+    test = Dataset.from_dict(train)
+    test = test.with_format("torch") # リスト型回避
+
+    # or
+    train_loader = DataLoader(train, batch_size=BATCH_SIZE, shuffle=True, drop_last=True)
+    test_loader = DataLoader(test, batch_size=BATCH_SIZE, shuffle=True, drop_last=True)
+    return train_loader, test_loader
 
 ```