Tensor data-related operation and function explanation _tensor

Tensor data-related operation and function explanation

Tensor

It is used in TensorFlow to represent data. Can be viewed as a multidimensional array or list.

Scalar is tensor, vector is tensor, matrix is tensor, matrix is tensor

Several definition methods commonly used

A 1.variable variable, typically a value that can be updated or changed, that is, the values that can be dynamically adjusted while the flow diagram is running. When we train a model, we use the variable (Variables) in TensorFlow, which we need to keep and update the parameter values, as well as the tensor, and the variables are stored in the memory buffer.

We have to initialize the variables beforehand, and the TensorFlow variables must be initialized before the values are available. The constant value tensor is not required, the variable can be set to initialize the way, but the real initialization is to Sess.run (Tf.global_variables_initializer ())

is really initialized.

2.CONSTANT constant tensor

3.placeholder: Placeholder Dynamic change value feeddict

NumPy

b = Np.array ([(1.5,2,3), (4,5,6)])

The difference between TensorFlow and NumPy

Same point: Both provide n-bit arrays

Different points: NumPy supports Ndarray, while TensorFlow has tensor;numpy does not provide create tensor functions and derivations, nor does it provide GPU support.

Show

Tensor

Need to add the Eval function

Ta = Tf.zeros ((2,2))

Print (TA)

Tensor ("zeros_1:0", Shape= (2, 2), Dtype=float32)

Print (Ta.eval ())

NumPy

A = Np.zeros ((2,2))

Print (a)

Tensor related Operations

Arithmetic operations

1. Addition operation

The effects of Tensor and NumPy two are automatically expanded when they encounter different dimensions uniformly. But the size of the same dimension must be the same, except for a certain dimension where the value is 1.

The shape of the tensor is (tensor,1) and (1,tensor) which can be added and automatically expanded.

2. Matrix multiplication

Tensor

A * B means to calculate by element

Tf.mul (A,B) represents a calculation by element

Tf.matmul (a,b) representation matrix multiplication

3.numpy

A * B means to calculate by element

Dot (a,b) representation matrix multiplication

Data type conversions

Tf.to_double (a)

Tf.to_float (a)

Tf.cast (x, Dtype, Name=none)

Tensor A is [1.8, 2.2], dtype=tf.float

Tf.cast (A, Tf.int32) ==> [1, 2] # Dtype=tf.int32

Shape action

1.shape

Numpy:a.shape ()

Tensor:a.get_shape () Tf.shape (a)

2.reshape

Tensor:tf.reshape (A, (1,4))

Numpy:np.reshape (A, (1,4))

3.tf.size (a) number of elements that return data

Tf.constant ([[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]]] size = 12

4.tf.rank (a) returns rank of tensor

# ' t ' is [[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]]

# shape of tensor ' t ' is [2, 2, 3]

Rank (t) ==> 3

5. A sum of one dimension

Tensor:tf.reduce_sum (B,reduction_indices=1)

Numpy:np.sum (B,axis=1)

Slices and merges for arrays

1. Merging, connecting arrays

Tensor

Tf.concat (0,[a,b]) The first argument expresses the number of digits

If a (1,128,128,3) b (1,128,128,3)

Tf.concat (0,[a,b]) (2,128,128,3)

NumPy

Vstack and Hstack

Stack (a,axis=)

2. Get whole row of data

Tensor

temp = Tf.constant (0,shape=[5,5])

Temp1 = temp[0,:] Getting a row

Temp2 = temp[:,1] Gets a column

temp[1,1] Get an element

Temp[1:3,1:3] Gets a range of row and column elements

Separation of tensor from Num_split tensors along a dimension

Tf.split (Split_dim, Num_split, value, name= ' Split ')

# ' value ' is a tensor with shape [5, 30]

# Split ' value ' into 3 tensors along Dimension 1

Split0, split1, split2 = Tf.split (1, 3, value)

Tf.shape (split0) ==> [5, 10]

3. Slice operations on tensor

Tf.slice (input_, begin, size, Name=none)

# ' input ' is

#[[[1, 1, 1], [2, 2, 2]],[[3, 3, 3], [4, 4, 4]],[[5, 5, 5], [6, 6, 6]]]

Tf.slice (input, [1, 0, 0], [1, 1, 3]) ==> [[[3, 3, 3]]]

Tf.slice (input, [1, 0, 0], [1, 2, 3]) ==>

[[[3, 3, 3],

[4, 4, 4]]]

Tf.slice (input, [1, 0, 0], [2, 1, 3]) ==>

[[[3, 3, 3]],

[[5, 5, 5]]]

4. Packing

Tf.pack (values, axis=0, name= ' pack ')

# ' x ' is [1, 4], ' Y ' are [2, 5], ' Z ' is [3, 6]

Pack ([x, Y, z]) => [[1, 4], [2, 5], [3, 6]]

# along the first dimension pack

Pack ([x, Y, z], Axis=1) => [[1, 2, 3], [4, 5, 6]]

Equivalent to Tf.pack ([x, Y, z]) = Np.asarray ([x, Y, z])

5.tf.reverse (tensor, dims, Name=none)

Sequence reversal along a dimension

Where Dim is a list, the element is bool, size equals rank (tensor)

# tensor ' t ' is

[[[[0, 1, 2, 3],

#[4, 5, 6, 7],

#[8, 9, 10, 11],

#[[12, 13, 14, 15],

#[16, 17, 18, 19],

#[20, 21, 22, 23]]]

# tensor ' t ' shape is [1, 2, 3, 4]

# ' dims ' is [False, False, False, True]

Reverse (t, dims) ==>

[[[[3, 2, 1, 0],

[7, 6, 5, 4],

[11, 10, 9, 8]],

[[15, 14, 13, 12],

[19, 18, 17, 16],

[23, 22, 21, 20]]]

6.tf.transpose (A, Perm=none, name= ' transpose ')

Swap tensor dimension Order

If defined, the perm is (n-1 ...). 0)

# ' x ' is [[1 2 3],[4 5 6]]

Tf.transpose (x) ==> [[1 4], [2 5],[3 6]]

# equivalently

Tf.transpose (x, perm=[1, 0]) ==> [[1 4],[2 5], [3 6]]

Matrix related Operations

Inverse matrix of 1.tf.matrix_inverse matrices

Determinant of 2.tf.matrix_determinant matrices

3.tf.transpose Transpose

4.tf.diag the value on the given diagonal to return diagonally tensor

Tensor and numpy Array Mutual transfer

1.numpy Array to Tensor

Numpydata = Np.zeros ((1,10,10,3), Dtype=np.float32)

Tf.convert_to_tensor (Numpydata)

2.Tensor to NumPy array

Eval ()

Tf.constant ([1,2,3]). Eval ()

Reference documents

http://blog.csdn.net/lenbow/article/details/52152766