Learning about tensor (III)

Index and slice

Index method ：

The first one is ： The most basic indexing method , The second kind ： use ： Go to index , The third kind of ： use ... Go to index .（ I don't know if it's right ）

First, let's talk about the first ：

a = tf.ones([1,5,5,3])
print(a[0])
print(a[0][0])
print(a[0][0][0])     # The most basic indexing method , Give the value of each layer 

Give the value of each layer .

The second kind ：

b = tf.random.normal([4,28,28,3])
print(b[1].shape)
print(b[1,2].shape)  # =b[1][2]


c=tf.range(10)# Use positive numbers near the initial segment , End with a negative number 
print(c[-1:])#  c[a:b]  from a Fetch b（ It doesn't contain b）, Don't write b Words represent from a To the end , Don't write a The words represent from the initial paragraph to b
# Don't write = Default 
# Take a piece of data 


#a[0].shape = a[0,:,:,:]
print(a[:,:,:,0].shape)  # Take the middle dimension 
print(a[:,0,:,:].shape)





# Add one more   :   Take across 
#a[::3] Take one data every three 

print(a[:,0::4,:,:]) # Take it every four points 

print(a[::-1]) # flashback 
print(a[::-2])  # Take across 

Based on the first one, the simplified operation is carried out , And achieve data retrieval across fixed data .

notes ：c[a:b] from a Fetch b（ It doesn't contain b）, Don't write b Words represent from a To the end , Don't write a The words represent from the initial paragraph to b

The third kind of ： Simplify the operation based on the second ：

# ...  # Automatic derivation 
# Instead of consecutive colons 

print(a[...,0])    #0 Represents the last dimension 
print(a[:,:,:,0])

Slicing :

#Selective indexing

utilize gather Slice it .

#Selective indexing

#gather   collect 
# Specify the number of lines 

d = tf.random.normal([4,35,8])
print(tf.gather(a,axis=0,indices=[2,3]))
#axis   Which dimension does it represent       indices Which index does it represent 
# Can be used in combination with 

 (axis=) Represents the dimension ,indices=  It represents the index

print(tf.gather_nd(d,[0]).shape)
# As d[0]
print(tf.gather_nd(d,[[[0,0,0],[1,1,1],[2,2,2]]]))
print(tf.gather_nd(d,[[0,0,0],[1,1,1],[2,2,2]]))

With one bracket and two brackets shape Will change .

utilize boolean_mask Slice it .

a=tf.random.normal([4,28,28,3])
print(tf.boolean_mask(a,mask=[True,True,False,False]).shape)
print(tf.boolean_mask(a,mask=[True,True,False],axis=3).shape)
# Not specified axis The default is 0

 

 

Note that... Is not specified axis The default is 0.

a = tf.ones([2,3,4])
print(tf.boolean_mask(a,mask=[[True,False,False],[False,True,True]]))

 

Slice the corresponding dimension according to the brackets in brackets .