CIFAR10. CODE 一些錯誤問題

cifar10 epoch validation_data=(x_val, y_val)) valueerror: input 0 of layer "sequential_1" is incompatible with the layer: expected shape=(none, 32, 32, 3), found shape=(none, 3072)

orga_an21 2022-06-21 16:24:10 ‧ 994 瀏覽

分享至

建立模型這裡開始並沒有錯

model = models.Sequential([
    layers.Conv2D(filters=32, kernel_size=(3, 3),activation="relu",input_shape=(32,32,3)),
    layers.MaxPooling2D((2, 2)),
    
    layers.Conv2D(filters=64, kernel_size=(3, 3),activation="relu"),
    layers.MaxPooling2D((2, 2)),
    
    layers.Flatten(),
    layers.Dense(100,activation="relu"),
    layers.Dense(10,activation="softmax")
    
    
    
    
])

再來到Training的地方就錯了，請問該如何修改

model.compile(loss=losses.categorical_crossentropy,
              optimizer=optimizers.SGD(),
              metrics=['accuracy']
              )

下面這步就錯了

logs = model.fit(x_train, y_train,
                batch_size=128,
                epochs=10,
                validation_data=(x_val, y_val))

再來後面的也提供給大家不知道後面的還有沒有錯誤

history = logs.history
print(history)

plt.plot(history['accuracy'])
plt.plot(history['val_accuracy'])
plt.legend(['accuracy', 'val_accuracy'])
plt.title('accuracy')

max_acc = np.max(history['val_accuracy'])
print(max_acc)

plt.plot(history['loss'])
plt.plot(history['val_loss'])
plt.legend(['loss', 'val_loss'])
plt.title('loss')

min_loss = np.min(history['val_loss'])
print(min_loss)

df = pd.DataFrame()
df['Id'] = [str(i) for i in range(len(x_test))]
predictions = model.predict(x_test)
predictions = np.argmax(predictions, axis=-1)
df['Category'] = predictions
df.to_csv('result.csv', index=None)
df

麻煩大家可以幫忙看看要怎麼修改才能做到最後,並存成一個csv檔

登入發表討論

直播研討會

{{ item.channelVendor }} {{ item.webinarstarted }} |

直播中

1 個回答

hokou

iT邦好手 1 級 ‧ 2022-06-21 16:31:01

input 錯了喔
設定(32,32,3)，結果進去的是(3072)

==
更新最後驗證狀況

加入 onehot 轉換
model 最後是分類數，所以需為10

x_train, x_val, y_train, y_val = train_test_split(x_train, y_train, test_size=0.2)

# 1. 加入 onehot 轉換
y_train = tf.keras.utils.to_categorical(y_train)
y_val = tf.keras.utils.to_categorical(y_val)
# ====

# 其他維持不動

model = models.Sequential([
    layers.Conv2D(filters=32, kernel_size=(3, 3),activation="relu",input_shape=(32,32,3)),
# 中間依原本設計，說明原因故刪除

# 2. 最後是分類數，所以需為10
    layers.Dense(10,activation="softmax")

])

這樣就可以訓練了

回應 9
分享
檢舉

看更多先前的回應...收起先前的回應...

orga_an21 iT邦新手 5 級 ‧ 2022-06-21 16:33:19 檢舉

那請問需要哪裡做處理，建立模型前面的CODE要提供給你看看嗎?

hokou iT邦好手 1 級 ‧ 2022-06-21 16:43:14 檢舉

你可以確認 x_train、x_val 單個資料的shape是不是(32,32,3)

orga_an21 iT邦新手 5 級 ‧ 2022-06-21 19:30:26 檢舉

我看好像沒有錯，還是我漏看了

import tensorflow as tf
from tensorflow.keras import datasets, models, layers, utils, activations, losses, optimizers, metrics
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd

tf.__version__

num_classes = 10
img_size = 32

!gdown --id '1RaH2V6FdhBxjUbREJMY2SdLS-O_Qhkq7' --output cifar10.npz

with np.load('cifar10.npz', allow_pickle=True) as f:
    x_train, y_train = f['x_train'], f['y_train']
    x_test = f['x_test']

x_train.shape, y_train.shape, x_test.shape

from sklearn.model_selection import train_test_split
x_train, x_val, y_train, y_val = train_test_split(x_train, y_train, test_size=0.2)

x_train.shape, x_val.shape, x_train.shape, x_val.shape

再來就是建立模型了

hokou iT邦好手 1 級 ‧ 2022-06-21 21:37:38 檢舉

以你截出來的shape，看起來是沒什麼問題
就要看你中間是否有再處理到什麼數值，導致變成(3072)了
剛發現 3072 是 32x32x3 降到 1維

orga_an21 iT邦新手 5 級 ‧ 2022-06-22 16:39:08 檢舉

model = models.Sequential([
    layers.Conv2D(filters=32, kernel_size=(3, 3),activation="relu",input_shape=(32,32,3)),
    layers.MaxPooling2D((2, 2)),
    
    layers.Conv2D(filters=64, kernel_size=(3, 3),activation="relu"),
    layers.MaxPooling2D((2, 2)),
    
    layers.Flatten(),
    layers.Dense(100,activation="relu"),
    layers.Dense(10,activation="softmax")
    
    
    
    
])

只有這裡的layers.Dense(100,activation="relu"), layers.Dense(10,activation="softmax")

100跟10有換過，我試過數值換成10跟20

出現
` ValueError: Shapes (None, 1) and (None, 20) are incompatible`

會是這關西嗎?

hokou iT邦好手 1 級 ‧ 2022-06-22 17:05:11 檢舉

最後一層是你的分類數，以這個資料集是10類來說要設定10

看是否是 label 未轉換的問題，把 label 轉 onehot

y_train = tf.keras.utils.to_categorical(y_train)
y_val = tf.keras.utils.to_categorical(y_val)

orga_an21 iT邦新手 5 級 ‧ 2022-06-22 17:14:13 檢舉

還是說我提供全部的code,給您看看，因為留言這樣感覺有點亂掉了

import tensorflow as tf
from tensorflow.keras import datasets, models, layers, utils, activations, losses, optimizers, metrics
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd

tf.__version__

num_classes = 10
img_size = 32

!gdown --id '1RaH2V6FdhBxjUbREJMY2SdLS-O_Qhkq7' --output cifar10.npz

with np.load('cifar10.npz', allow_pickle=True) as f:
    x_train, y_train = f['x_train'], f['y_train']
    x_test = f['x_test']

x_train.shape, y_train.shape, x_test.shape

from sklearn.model_selection import train_test_split
x_train, x_val, y_train, y_val = train_test_split(x_train, y_train, test_size=0.2)

x_train.shape, x_val.shape, x_train.shape, x_val.shape

model = models.Sequential([
    layers.Conv2D(filters=32, kernel_size=(3, 3),activation="relu",input_shape=(32,32,3)),
    layers.MaxPooling2D((2, 2)),
    
    layers.Conv2D(filters=64, kernel_size=(3, 3),activation="relu"),
    layers.MaxPooling2D((2, 2)),
    
    layers.Flatten(),
    layers.Dense(10,activation="relu"),
    layers.Dense(20,activation="softmax")
    
    
    
    
])

model.compile(loss=losses.categorical_crossentropy,
              optimizer=optimizers.SGD(),
              metrics=['accuracy']
              )

logs = model.fit(x_train, y_train,
                batch_size=128,
                epochs=10)
                #validation_data=(x_val, y_val))

history = logs.history
print(history)

plt.plot(history['accuracy'])
plt.plot(history['val_accuracy'])
plt.legend(['accuracy', 'val_accuracy'])
plt.title('accuracy')

max_acc = np.max(history['val_accuracy'])
print(max_acc)

plt.plot(history['loss'])
plt.plot(history['val_loss'])
plt.legend(['loss', 'val_loss'])
plt.title('loss')

min_loss = np.min(history['val_loss'])
print(min_loss)

df = pd.DataFrame()
df['Id'] = [str(i) for i in range(len(x_test))]
predictions = model.predict(x_test)
predictions = np.argmax(predictions, axis=-1)
df['Category'] = predictions
df.to_csv('result.csv', index=None)
df

這是全部了，如有時間願意的話再請幫幫忙，感謝，我實在是找不出問題

還是說我提供全部的code,給您看看，因為留言這樣感覺有點亂掉了
```
import tensorflow as tf
from tensorflow.keras import datasets, models, layers, utils, activations, losses, optimizers, metrics
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
```

```
tf.__version__
```

```
num_classes = 10
img_size = 32
```

```
!gdown --id '1RaH2V6FdhBxjUbREJMY2SdLS-O_Qhkq7' --output cifar10.npz
```

```
with np.load('cifar10.npz', allow_pickle=True) as f:
    x_train, y_train = f['x_train'], f['y_train']
    x_test = f['x_test']

x_train.shape, y_train.shape, x_test.shape
```

```
from sklearn.model_selection import train_test_split
x_train, x_val, y_train, y_val = train_test_split(x_train, y_train, test_size=0.2)

x_train.shape, x_val.shape, x_train.shape, x_val.shape
```

```
model = models.Sequential([
    layers.Conv2D(filters=32, kernel_size=(3, 3),activation="relu",input_shape=(32,32,3)),
    layers.MaxPooling2D((2, 2)),
    
    layers.Conv2D(filters=64, kernel_size=(3, 3),activation="relu"),
    layers.MaxPooling2D((2, 2)),
    
    layers.Flatten(),
    layers.Dense(10,activation="relu"),
    layers.Dense(20,activation="softmax")
    
    
    
    
])
```

```
model.compile(loss=losses.categorical_crossentropy,
              optimizer=optimizers.SGD(),
              metrics=['accuracy']
              )
```

```
logs = model.fit(x_train, y_train,
                batch_size=128,
                epochs=10)
                #validation_data=(x_val, y_val))
```

```
history = logs.history
print(history)
```

```
plt.plot(history['accuracy'])
plt.plot(history['val_accuracy'])
plt.legend(['accuracy', 'val_accuracy'])
plt.title('accuracy')
```

```
max_acc = np.max(history['val_accuracy'])
print(max_acc)
```

```
plt.plot(history['loss'])
plt.plot(history['val_loss'])
plt.legend(['loss', 'val_loss'])
plt.title('loss')
```

```
min_loss = np.min(history['val_loss'])
print(min_loss)
```

```
df = pd.DataFrame()
df['Id'] = [str(i) for i in range(len(x_test))]
predictions = model.predict(x_test)
predictions = np.argmax(predictions, axis=-1)
df['Category'] = predictions
df.to_csv('result.csv', index=None)
df
```

這是全部了，如有時間願意的話再請幫幫忙，感謝，我實在是找不出問題

修改

hokou iT邦好手 1 級 ‧ 2022-06-23 09:09:19 檢舉

x_train, x_val, y_train, y_val = train_test_split(x_train, y_train, test_size=0.2)

# 1. 加入 onehot 轉換
y_train = tf.keras.utils.to_categorical(y_train)
y_val = tf.keras.utils.to_categorical(y_val)
# ====

# 其他維持不動

model = models.Sequential([
    layers.Conv2D(filters=32, kernel_size=(3, 3),activation="relu",input_shape=(32,32,3)),
# 中間依原本設計，說明原因故刪除

# 2. 最後是分類數，所以需為10
    layers.Dense(10,activation="softmax")

])

這樣就可以跑了喔

後續的 plt 和 df 我沒確認