# TensorFlow中常用函数说明

### 2、tf函数

Maths Add, Sub, Mul, Div, Exp, Log, Greater, Less, Equal
Array Concat, Slice, Split, Constant, Rank, Shape, Shuffle
Matrix MatMul, MatrixInverse, MatrixDeterminant
Neuronal Network SoftMax, Sigmoid, ReLU, Convolution2D, MaxPool
Checkpointing Save, Restore
Queues and syncronizations Enqueue, Dequeue, MutexAcquire, MutexRelease
Flow control Merge, Switch, Enter, Leave, NextIteration

#### TensorFlow的算术操作如下：

tf.sub(x, y, name=None) 减法
tf.mul(x, y, name=None) 乘法
tf.div(x, y, name=None) 除法
tf.mod(x, y, name=None) 取模
tf.abs(x, name=None) 求绝对值
tf.neg(x, name=None) 取负 (y = -x).
tf.sign(x, name=None) 返回符号 y = sign(x) = -1 if x < 0; 0 if x == 0; 1 if x > 0.
tf.inv(x, name=None) 取反
tf.square(x, name=None) 计算平方 (y = x * x = x^2).
tf.round(x, name=None) 舍入最接近的整数
# ‘a’ is [0.9, 2.5, 2.3, -4.4]
tf.round(a) ==> [ 1.0, 3.0, 2.0, -4.0 ]
tf.sqrt(x, name=None) 开根号 (y = \sqrt{x} = x^{1/2}).
tf.pow(x, y, name=None) 幂次方
# tensor ‘x’ is [[2, 2], [3, 3]]
# tensor ‘y’ is [[8, 16], [2, 3]]
tf.pow(x, y) ==> [[256, 65536], [9, 27]]
tf.exp(x, name=None) 计算e的次方
tf.log(x, name=None) 计算log，一个输入计算e的ln，两输入以第二输入为底
tf.maximum(x, y, name=None) 返回最大值 (x > y ? x : y)
tf.minimum(x, y, name=None) 返回最小值 (x < y ? x : y)
tf.cos(x, name=None) 三角函数cosine
tf.sin(x, name=None) 三角函数sine
tf.tan(x, name=None) 三角函数tan
tf.atan(x, name=None) 三角函数ctan

#### 张量操作Tensor Transformations

• 数据类型转换Casting

tf.string_to_number
(string_tensor, out_type=None, name=None)

tf.to_double(x, name=’ToDouble’) 转为64位浮点类型–float64
tf.to_float(x, name=’ToFloat’) 转为32位浮点类型–float32
tf.to_int32(x, name=’ToInt32’) 转为32位整型–int32
tf.to_int64(x, name=’ToInt64’) 转为64位整型–int64
tf.cast(x, dtype, name=None) 将x或者x.values转换为dtype
# tensor a is [1.8, 2.2], dtype=tf.float
tf.cast(a, tf.int32) ==> [1, 2] # dtype=tf.int32

• 形状操作Shapes and Shaping

tf.shape(input, name=None) 返回数据的shape
# ‘t’ is [[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]]
shape(t) ==> [2, 2, 3]
tf.size(input, name=None) 返回数据的元素数量
# ‘t’ is [[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]]]
size(t) ==> 12
tf.rank(input, name=None) 返回tensor的rank

tensor的rank表示一个tensor需要的索引数目来唯一表示任何一个元素

#’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
tf.reshape(tensor, shape, name=None) 改变tensor的形状
# tensor ‘t’ is [1, 2, 3, 4, 5, 6, 7, 8, 9]
# tensor ‘t’ has shape [9]
reshape(t, [3, 3]) ==>
[[1, 2, 3],
[4, 5, 6],
[7, 8, 9]]
#如果shape有元素[-1],表示在该维度打平至一维
# -1 将自动推导得为 9:
reshape(t, [2, -1]) ==>
[[1, 1, 1, 2, 2, 2, 3, 3, 3],
[4, 4, 4, 5, 5, 5, 6, 6, 6]]
tf.expand_dims(input, dim, name=None) 插入维度1进入一个tensor中
#该操作要求-1-input.dims()
# ‘t’ is a tensor of shape [2]
shape(expand_dims(t, 0)) ==> [1, 2]
shape(expand_dims(t, 1)) ==> [2, 1]
shape(expand_dims(t, -1)) ==> [2, 1] <= dim <= input.dims()
• 切片与合并（Slicing and Joining）

tf.slice(input_, begin, size, name=None) 对tensor进行切片操作

#’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]]]
tf.split(split_dim, num_split, value, name=’split’) 沿着某一维度将tensor分离为num_split tensors
# ‘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]
tf.concat(concat_dim, values, name=’concat’) 沿着某一维度连结tensor
t1 = [[1, 2, 3], [4, 5, 6]]
t2 = [[7, 8, 9], [10, 11, 12]]
tf.concat(0, [t1, t2]) ==> [[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12]]
tf.concat(1, [t1, t2]) ==> [[1, 2, 3, 7, 8, 9], [4, 5, 6, 10, 11, 12]]

tf.concat(axis, [tf.expand_dims(t, axis) for t in tensors])

tf.pack(values, axis=0, name=’pack’) 将一系列rank-R的tensor打包为一个rank-(R+1)的tensor
# ‘x’ is [1, 4], ‘y’ is [2, 5], ‘z’ is [3, 6]
pack([x, y, z]) => [[1, 4], [2, 5], [3, 6]]
# 沿着第一维pack
pack([x, y, z], axis=1) => [[1, 2, 3], [4, 5, 6]]

tf.reverse(tensor, dims, name=None) 沿着某维度进行序列反转

# 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]]]]
tf.transpose(a, perm=None, name=’transpose’) 调换tensor的维度顺序

# ‘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]]
tf.gather(params, indices, validate_indices=None, name=None) 合并索引indices所指示params中的切片
tf.one_hot
(indices, depth, on_value=None, off_value=None,
axis=None, dtype=None, name=None)
indices = [0, 2, -1, 1]
depth = 3
on_value = 5.0
off_value = 0.0
axis = -1
#Then output is [4 x 3]:
output =
[5.0 0.0 0.0] // one_hot(0)
[0.0 0.0 5.0] // one_hot(2)
[0.0 0.0 0.0] // one_hot(-1)
[0.0 5.0 0.0] // one_hot(1)

#### 矩阵相关运算

tf.diag(diagonal, name=None) 返回一个给定对角值的对角tensor
# ‘diagonal’ is [1, 2, 3, 4]
tf.diag(diagonal) ==>
[[1, 0, 0, 0]
[0, 2, 0, 0]
[0, 0, 3, 0]
[0, 0, 0, 4]]
tf.diag_part(input, name=None) 功能与上面相反
tf.trace(x, name=None) 求一个2维tensor足迹，即对角值diagonal之和
tf.transpose(a, perm=None, name=’transpose’) 调换tensor的维度顺序

# ‘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]]
tf.matmul(a, b, transpose_a=False,
transpose_b=False, a_is_sparse=False,
b_is_sparse=False, name=None)

tf.matrix_determinant(input, name=None) 返回方阵的行列式
tf.cholesky(input, name=None) 对输入方阵cholesky分解，

matrix为方阵shape为[M,M],rhs的shape为[M,K]，output为[M,K]

#### 复数操作

tf.complex(real, imag, name=None) 将两实数转换为复数形式
# tensor ‘real’ is [2.25, 3.25]
# tensor imag is [4.75, 5.75]
tf.complex(real, imag) ==> [[2.25 + 4.75j], [3.25 + 5.75j]]
tf.complex_abs(x, name=None) 计算复数的绝对值，即长度。
# tensor ‘x’ is [[-2.25 + 4.75j], [-3.25 + 5.75j]]
tf.complex_abs(x) ==> [5.25594902, 6.60492229]
tf.conj(input, name=None) 计算共轭复数
tf.imag(input, name=None)
tf.real(input, name=None)

tf.fft(input, name=None) 计算一维的离散傅里叶变换，输入数据类型为complex64

#### 归约计算(Reduction)

tf.reduce_sum(input_tensor, reduction_indices=None,
keep_dims=False, name=None)

# ‘x’ is [[1, 1, 1]
# [1, 1, 1]]
tf.reduce_sum(x) ==> 6
tf.reduce_sum(x, 0) ==> [2, 2, 2]
tf.reduce_sum(x, 1) ==> [3, 3]
tf.reduce_sum(x, 1, keep_dims=True) ==> [[3], [3]]
tf.reduce_sum(x, [0, 1]) ==> 6
tf.reduce_prod(input_tensor,
reduction_indices=None,
keep_dims=False, name=None)

tf.reduce_min(input_tensor,
reduction_indices=None,
keep_dims=False, name=None)

tf.reduce_max(input_tensor,
reduction_indices=None,
keep_dims=False, name=None)

tf.reduce_mean(input_tensor,
reduction_indices=None,
keep_dims=False, name=None)

tf.reduce_all(input_tensor,
reduction_indices=None,
keep_dims=False, name=None)

# ‘x’ is
# [[True, True]
# [False, False]]
tf.reduce_all(x) ==> False
tf.reduce_all(x, 0) ==> [False, False]
tf.reduce_all(x, 1) ==> [True, False]
tf.reduce_any(input_tensor,
reduction_indices=None,
keep_dims=False, name=None)

tf.accumulate_n(inputs, shape=None,
tensor_dtype=None, name=None)

# tensor ‘a’ is [[1, 2], [3, 4]]
# tensor b is [[5, 0], [0, 6]]
tf.accumulate_n([a, b, a]) ==> [[7, 4], [6, 14]]
tf.cumsum(x, axis=0, exclusive=False,
reverse=False, name=None)

tf.cumsum([a, b, c]) ==> [a, a + b, a + b + c]
tf.cumsum([a, b, c], exclusive=True) ==> [0, a, a + b]
tf.cumsum([a, b, c], reverse=True) ==> [a + b + c, b + c, c]
tf.cumsum([a, b, c], exclusive=True, reverse=True) ==> [b + c, c, 0]

#### 分割(Segmentation)

tf.segment_sum(data, segment_ids, name=None) 根据segment_ids的分段计算各个片段的和

c = tf.constant([[1,2,3,4], [-1,-2,-3,-4], [5,6,7,8]])
tf.segment_sum(c, tf.constant([0, 0, 1]))
==>[[0 0 0 0]
[5 6 7 8]]

tf.segment_prod(data, segment_ids, name=None) 根据segment_ids的分段计算各个片段的积
tf.segment_min(data, segment_ids, name=None) 根据segment_ids的分段计算各个片段的最小值
tf.segment_max(data, segment_ids, name=None) 根据segment_ids的分段计算各个片段的最大值
tf.segment_mean(data, segment_ids, name=None) 根据segment_ids的分段计算各个片段的平均值
tf.unsorted_segment_sum(data, segment_ids,
num_segments, name=None)

tf.sparse_segment_sum(data, indices,
segment_ids, name=None)

c = tf.constant([[1,2,3,4], [-1,-2,-3,-4], [5,6,7,8]])
# Select two rows, one segment.
tf.sparse_segment_sum(c, tf.constant([0, 1]), tf.constant([0, 0]))
==> [[0 0 0 0]]

#### 序列比较与索引提取(Sequence Comparison and Indexing)

tf.argmin(input, dimension, name=None) 返回input最小值的索引index
tf.argmax(input, dimension, name=None) 返回input最大值的索引index
tf.listdiff(x, y, name=None) 返回x，y中不同值的索引
tf.where(input, name=None) 返回bool型tensor中为True的位置
# ‘input’ tensor is
#[[True, False]
#[True, False]]
# ‘input’ 有两个’True’,那么输出两个坐标值.
# ‘input’的rank为2, 所以每个坐标为具有两个维度.
where(input) ==>
[[0, 0],
[1, 0]]
tf.unique(x, name=None) 返回一个元组tuple(y,idx)，y为x的列表的唯一化数据列表，
idx为x数据对应y元素的index
# tensor ‘x’ is [1, 1, 2, 4, 4, 4, 7, 8, 8]
y, idx = unique(x)
y ==> [1, 2, 4, 7, 8]
idx ==> [0, 0, 1, 2, 2, 2, 3, 4, 4]
tf.invert_permutation(x, name=None) 置换x数据与索引的关系
# tensor x is [3, 4, 0, 2, 1]
invert_permutation(x) ==> [2, 4, 3, 0, 1]

#### 神经网络(Neural Network)

• 激活函数（Activation Functions）

tf.nn.relu(features, name=None) 整流函数：max(features, 0)
tf.nn.relu6(features, name=None) 以6为阈值的整流函数：min(max(features, 0), 6)
tf.nn.elu(features, name=None) elu函数，exp(features) - 1 if < 0,否则features
Exponential Linear Units (ELUs)
tf.nn.softplus(features, name=None) 计算softplus：log(exp(features) + 1)
tf.nn.dropout(x, keep_prob,
noise_shape=None, seed=None, name=None)

noise_shape为噪声的shape

tf.sigmoid(x, name=None) y = 1 / (1 + exp(-x))
tf.tanh(x, name=None) 双曲线切线激活函数
• 卷积函数（Convolution）

use_cudnn_on_gpu=None, data_format=None, name=None)

tf.nn.conv3d(input, filter, strides, padding, name=None) 在给定的5D input与 filter下计算3D卷积

• 池化函数（Pooling）

data_format=’NHWC’, name=None)

data_format=’NHWC’, name=None)

tf.nn.max_pool_with_argmax(input, ksize, strides,

tf.nn.avg_pool3d(input, ksize, strides,
3D平均值pooling
tf.nn.max_pool3d(input, ksize, strides,
3D最大值pooling
• 数据标准化（Normalization）

tf.nn.l2_normalize(x, dim, epsilon=1e-12, name=None) 对维度dim进行L2范式标准化
output = x / sqrt(max(sum(x**2), epsilon))
tf.nn.sufficient_statistics(x, axes, shift=None,
keep_dims=False, name=None)

tf.nn.normalize_moments(counts, mean_ss, variance_ss, shift, name=None) 基于完全统计量计算均值和方差
tf.nn.moments(x, axes, shift=None,
name=None, keep_dims=False)

• 损失函数（Losses）

tf.nn.l2_loss(t, name=None) output = sum(t ** 2) / 2
• 分类函数（Classification）

tf.nn.sigmoid_cross_entropy_with_logits
(logits, targets, name=None)*

tf.nn.softmax(logits, name=None) 计算softmax
softmax[i, j] = exp(logits[i, j]) / sum_j(exp(logits[i, j]))
tf.nn.log_softmax(logits, name=None) logsoftmax[i, j] = logits[i, j] - log(sum(exp(logits[i])))
tf.nn.softmax_cross_entropy_with_logits
(logits, labels, name=None)

logits, labels必须为相同的shape与数据类型
tf.nn.sparse_softmax_cross_entropy_with_logits
(logits, labels, name=None)

tf.nn.weighted_cross_entropy_with_logits
(logits, targets, pos_weight, name=None)

• 符号嵌入（Embeddings）

tf.nn.embedding_lookup
(params, ids, partition_strategy=’mod’,
name=None, validate_indices=True)

1、如果partition_strategy为”mod”，
id所分配到的位置为p = id % len(params)

[[0, 5, 10], [1, 6, 11], [2, 7, 12], [3, 8], [4, 9]]
2、如果partition_strategy为”div”,那么分配方案为：
[[0, 1, 2], [3, 4, 5], [6, 7, 8], [9, 10], [11, 12]]
tf.nn.embedding_lookup_sparse(params,
sp_ids, sp_weights, partition_strategy=’mod’,
name=None, combiner=’mean’)

1、sp_ids为一个N x M的稀疏tensor，
N为batch大小，M为任意，数据类型int64
2、sp_weights的shape与sp_ids的稀疏tensor权重，

• 循环神经网络（Recurrent Neural Networks）

tf.nn.rnn(cell, inputs, initial_state=None, dtype=None,
sequence_length=None, scope=None)

tf.nn.dynamic_rnn(cell, inputs, sequence_length=None,
initial_state=None, dtype=None, parallel_iterations=None,
swap_memory=False, time_major=False, scope=None)

tf.nn.state_saving_rnn(cell, inputs, state_saver, state_name,
sequence_length=None, scope=None)

tf.nn.bidirectional_rnn(cell_fw, cell_bw, inputs,
initial_state_fw=None, initial_state_bw=None, dtype=None,
sequence_length=None, scope=None)

(outputs, output_state_fw, output_state_bw)
• 求值网络（Evaluation）

tf.nn.top_k(input, k=1, sorted=True, name=None) 返回前k大的值及其对应的索引
tf.nn.in_top_k(predictions, targets, k, name=None) 返回判断是否targets索引的predictions相应的值

Sampled Loss Functions
tf.nn.nce_loss(weights, biases, inputs, labels, num_sampled,
num_classes, num_true=1, sampled_values=None,
remove_accidental_hits=False, partition_strategy=’mod’,
name=’nce_loss’)

tf.nn.sampled_softmax_loss(weights, biases, inputs, labels,
num_sampled, num_classes, num_true=1, sampled_values=None,
remove_accidental_hits=True, partition_strategy=’mod’,
name=’sampled_softmax_loss’)

Candidate Samplers
tf.nn.uniform_candidate_sampler(true_classes, num_true,
num_sampled, unique, range_max, seed=None, name=None)

1、sampled_candidates 候选集合。
2、期望的true_classes个数，为浮点值
3、期望的sampled_candidates个数，为浮点值
tf.nn.log_uniform_candidate_sampler(true_classes, num_true,
num_sampled, unique, range_max, seed=None, name=None)

tf.nn.learned_unigram_candidate_sampler
(true_classes, num_true, num_sampled, unique,
range_max, seed=None, name=None)

tf.nn.fixed_unigram_candidate_sampler(true_classes, num_true,
num_sampled, unique, range_max, vocab_file=”,
distortion=1.0, num_reserved_ids=0, num_shards=1,
shard=0, unigrams=(), seed=None, name=None)

# 保存与恢复变量

tf.train.Saver.__init__(var_list=None, reshape=False,
sharded=False, max_to_keep=5,
keep_checkpoint_every_n_hours=10000.0,
name=None, restore_sequentially=False,
saver_def=None, builder=None)

var_list定义需要存储和恢复的变量
tf.train.Saver.save(sess, save_path, global_step=None,
latest_filename=None, meta_graph_suffix=’meta’,
write_meta_graph=True)

tf.train.Saver.restore(sess, save_path) 恢复变量
tf.train.Saver.last_checkpoints 列出最近未删除的checkpoint 文件名
tf.train.Saver.set_last_checkpoints(last_checkpoints) 设置checkpoint文件名列表
tf.train.Saver.set_last_checkpoints_with_time(last_checkpoints_with_time) 设置checkpoint文件名列表和时间戳

• 广告
• 抄袭
• 版权
• 政治
• 色情
• 无意义
• 其他

120