1. tensor
Tensor can be said to mark TensorFlow, because the name of the entire framework TensorFlow is meant tensor current, comprehensive understanding of what tensor. In TensorFlow tensor program uses data structure to represent all of the data in the computation graph, between the operating data are Tensor, Tensor can be seen as the n-dimensional array or list, each tensor contains the type (type), order ( rank), and shape (shape).
2.tensor type
Type tensor are mainly as follows:
tf.float32: 32-bit floating point
tf.float64: 64-bit floating point
tf.int64: 64-bit signed integer
tf.int32: 32-bit signed integer
tf.int16: 16-bit signed integer
tf. int8: 8-bit signed integer
tf.uint8: 8-bit unsigned integer
tf.string: variable length byte array
tf.bool: Boolean
tf.complex64: a plurality of two 32-bit floating-point consisting of: real and imaginary portions
rank (rank)
rank (order) refers to the dimension, which can be observed by the parentheses the number of layers, such as tensor [[1,2,3], [2,3,4], [3,4,5]] of the order of 2 order, scalar, vector, matrix 0,1,2 respectively.
shape (shape)
used to describe the shape of the internal organizational relationship tensor, the shape usually be a list of integers or tuples that can also be used in the correlation TensorFlow shape function to represent.
Tensor related operations
Tensor related operations including the type of operation, the digital operation, shape conversion, data manipulation
Type of operation
IF __name__ == ' __main__ ' : with tf.Session () AS sess: Print (sess.run (tf.string_to_number ( ' 123.456 ' ))) # characters into digital Print (sess.run (tf.to_double (3 ))) # turn floating- Print (sess.run (tf.to_int32 (3.1415))) # turn integer Print (sess.run (tf.cast (3.1415, tf.int32))) # the type of transfer specified type
Import tensorflow TF AS IF the __name__ == ' __main__ ' : with tf.Session () AS Sess: Print (sess.run (tf.ones ([2,. 3], DTYPE = tf.float32))) # generate a full 1 data Print (sess.run (tf.zeros ([2,. 3], DTYPE = tf.float32))) # generates data of all 0 Print (sess.run (tf.ones_like ([. 1, 2,. 3,. 4] ))) # generate a predetermined shape of the whole data Print (sess.run (tf.zeros_like ([2,. 3,. 4,. 5]))) # generated all-0 data of a predetermined shape Print (sess.run (tf.fill ([2, 2], 6))) # specified value fill shapes Print (sess.run (tf.constant ((2,. 3),. 3))) # generates constant print (sess.run (tf.random_normal ([3, 3], mean = 2.5, stddev = 1.0, dtype = tf.float32,)))
Too much. . . .