b.神經網路的資料表示法：張量Tensor

# b.神經網路的資料表示法：張量Tensor ###### tags: `Deep Learning by Keras God` > **#結論：將多個3維張量放到陣列，會成為4維張量** > **#同理：將多個4維張量放到陣列，會成為5維張量** ## 1.張量的關鍵屬性 * **匯入手寫辨識資料集(Mnist)** ```python=+ from keras.datasets import mnist (train_img,train_label),(test_img,test_label) = mnist.load_data() ``` > ```11493376/11490434 [==============================] - 0s 0us/step``` ## * **瞭解資料** ```python=+ print(train_img.ndim,train_img.shape,train_img.dtype) ``` > ```3 (60000, 28, 28) uint8``` ## * **查看圖片** ```python=+ import matplotlib.pyplot as plt digit = train_img[4] #查看第5張圖片 plt.imshow(digit,cmap = plt.cm.binary) plt.show() ``` > ![](https://i.imgur.com/tjswIZj.png) ## * **擷取特定資料** ```python=+ my_slice = train_img[10:100] print(my_slice.shape) my_slice = train_img[10:100,:,:] print(my_slice.shape) my_slice = train_img[10:100,0:28,0:28] print(my_slice.shape) ``` > ```(90, 28, 28)``` > ```(90, 28, 28)``` > ```(90, 28, 28)``` ## * **擷取圖片特定範圍** ```python=+ #切出每張圖片右下角14*14的像素 my_slice = train_img[:,14:,14:] print(my_slice.shape) #擷取每張圖片居中的14*14像素 my_slice = train_img[:,7:-7,7:-7] print(my_slice.shape) ``` > ```(60000, 14, 14)``` > ```(60000, 14, 14)``` ## 2.常見的資料型態 | 維度 | 型態 | |:----:|:------------------:| | #2D | 向量資料 | | #3D | 時間序列或序列資料 | | #4D | 影像 | | #5D | 視訊 | ## 3.神經網路的工具：張量運算 * **張量擴張 Broadcasting** ```python=+ import numpy as np x = np.random.random((64,3,32,10)) y = np.random.random((32,10)) #會自動擴張，但結尾維度數量要一樣 z = np.maximum(x,y) z.shape ``` > ```(64, 3, 32, 10)``` # * **張量擴張 Broadcasting** ```python=+ import numpy as np x = np.random.random((64,3,32,10)) y = np.random.random((32,10)) #會自動擴張，但結尾維度數量要一樣 z = np.maximum(x,y) z.shape ``` > ```(64, 3, 32, 10)``` # * **張量點積運算 Dot** ```python=+ import numpy as np x = np.random.random((64,3,32,10)) y = np.random.random((32,10)) #會自動擴張，但結尾維度數量要一樣 z = np.maximum(x,y) z.shape ``` > ```(64, 3, 32, 10)``` ## 時間戳記 > [name=ZEOxO][time=Sun, Dec 06, 2020 15:53 PM][color=#907bf7]