First look at what used to merge and split the interface tensor:
tf.concat tensor used for splicing, tf.stack used for stacking tensor, tensor tf.split used for segmentation, tf.unstack a tf.split also used for tensor segmentation
1.tf.concat
Axis represents the parameter that will be merged dimensions
# Assume that a score representative of four classes (class 35, 8 subjects), b represent 2 classes score a tf.ones = ([4,35,8 ]) B = tf.ones ([2,35,8 ]) # use concat merged results obtained six classes C = tf.concat ([A, B], Axis = 0) # (6,35,8) Print (c.shape)
2.tf.stack (for creating a new dimension)
# Assume that a score of four classes (class 35, 8 subjects) represents A school, b B represents school accomplishments four classes a tf.ones = ([4,35,8 ]) B = tf.ones ([4,35,8 ]) # using stack merged results obtained six classes C = tf.stack ([A, B], Axis = 0) # (2,4,35,8 ) Print (c.shape)
3.tf.unstack (aliquoted to a dimension)
# Assume that a score of four classes (class 35, 8 subjects) represents A school, b B represents school accomplishments four classes a tf.ones = ([4,35,8 ]) B = tf.ones ([4,35,8 ]) # using stack merged results obtained six classes C = tf.stack ([A, B], Axis = 0) # (2,4,35,8 ) Print (c.shape) aa,bb=tf.unstack(c,axis=0) # (4,35,8) print(aa.shape,bb.shape) res=tf.unstack(c,axis=3) # (2,4,35) print(res[0].shape,res[7].shape)
4.tf.split (scale break)
# Assume that a score of four classes (class 35, 8 subjects) represents A school, b B represents school accomplishments four classes a tf.ones = ([4,35,8 ]) B = tf.ones ([4,35,8 ]) # using stack merged results obtained six classes C = tf.stack ([A, B], Axis = 0) # (2,4,35,8 ) Print (c.shape) res = tf.split(c,axis=3,num_or_size_splits=2) # 2,(2,4,35,4) print(len(res),res[0].shape,res[1].shape) res = tf.split(c,axis=3,num_or_size_splits=[2,2,4]) # 3 (2,4,35,2) (2,4,35,2) (2,4,35,4) print(len(res),res[0].shape,res[1].shape,res[2].shape)