o Qe@s ddlmZddlmZddlmZddlmZddlmZddlm Z ddl Z dd l m Z dd l mZdd l mZmZmZdd lmZmZddlZdd lmZmZmZddlmZddlmZmZgZd[ddZd[ddZed[ddZd\ddZ d]ddZ!d^ddZ"d_d!d"Z#d`d#d$Z$dad&d'Z%dbd)d*Z&dcd.d/Z'd`d0d1Z(ed`d2d3Z)d`d4d5Z*ddd7d8Z+d`d9d:Z, ; < ded=d>Z-d`d?d@Z.dfdBdCZ/edfdDdEZ0dgdGdHZ1d]dIdJZ2d`dKdLZ3d`dMdNZ4d`dOdPZ5d`dQdRZ6d[dSdTZ7dfdUdVZ8dhdWdXZ9didYdZZ:dS)j)sigmoid)tanh)tanh_)inplace_apis_in_dygraph_only)chunk)multiplyN) LayerHelper)convert_np_dtype_to_dtype_)_in_legacy_dygraphin_dygraph_mode_non_static_mode)check_variable_and_dtype check_dtype)_C_ops _legacy_C_opsin_dynamic_mode)core)r r ?cCs|dkrtdtrt|d|Strt||St|dgddtd it}| |j }|j dd|id|id|id |S) a celu activation. .. math:: celu(x) = max(0, x) + min(0, \alpha * (e^{x/\alpha}-1)) Parameters: x (Tensor): The input Tensor with data type float32, float64. alpha (float, optional): The 'alpha' value of the CELU formulation. Default is 1.0. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([[-1., 6.], [1., 15.6]]) out = F.celu(x, alpha=0.2) # [[-0.19865242, 6. ], # [ 1. , 15.60000038]] rzalpha cannot be 0 for celualphaxZfloat16float32float64celuXOuttypeinputsoutputsattrsN)r) ZeroDivisionErrorr rrr rrr locals"create_variable_for_type_inferencedtype append_oprrnamehelperoutr+OD:\Projects\ConvertPro\env\Lib\site-packages\paddle/nn/functional/activation.pyr(s   rcCvtr t||Strt|d|St|dgddtd it}||j }|j dd|id|id|id|S) a elu activation. .. math:: elu(x)= \left\{ \begin{array}{lcl} x,& &\text{if } \ x > 0 \\ alpha * (e^{x} - 1),& &\text{if } \ x <= 0 \end{array} \right. Parameters: x (Tensor): The input Tensor with data type float32, float64. alpha (float, optional): The 'alpha' value of the ELU formulation. Default is 1.0. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([[-1., 6.], [1., 15.6]]) out = F.elu(x, alpha=0.2) # [[-0.12642411 6. ] # [ 1. 15.6 ]] rrrelurrrN)r.) r rr.r rrr r#r$r%r&r'r+r+r,r.Vs"  r.cCs0|dksJdtrt||St|d|S)z Inplace version of ``elu`` API, the output Tensor will be inplaced with input ``x``. Please refer to :ref:`api_nn_cn_elu`. gz5elu_ only support alpha >= 0, please use elu instead.r)r relu_r)rrr(r+r+r,r/s r/FcCr-) a gelu activation. The activation function of Gelu is calculated element by element. More information refers to :ref: `Gaussian Error Linear Units`. if approximate is True .. math:: gelu(x) = 0.5 * x * (1 + tanh(\sqrt{\frac{2}{\pi}} * (x + 0.044715x^{3}))) else .. math:: gelu(x) = 0.5 * x * (1 + erf(\frac{x}{\sqrt{2}})) Parameters: x (Tensor): The input Tensor with data type float32, float64. approximate (bool, optional): Whether to enable approximation. Default is False. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([[-1, 0.5], [1, 1.5]]) out1 = F.gelu(x) # [[-0.15865529, 0.34573123], # [ 0.84134471, 1.39978933]] out2 = F.gelu(x, True) # [[-0.15880799, 0.34571400], # [ 0.84119201, 1.39957154]] approximaterrgelurrrN)r1) r rr1r rrr r#r$r%r&)rr0r(r)r*r+r+r,r1s)  r1?cCsvtr t||Strt|d|St|dgddtd it}||j }|j dd|id|id|id|S) a hard shrinkage activation .. math:: hardshrink(x)= \left\{ \begin{array}{rcl} x,& &if \ {x > threshold} \\ x,& &if \ {x < -threshold} \\ 0,& &if \ {others} & \end{array} \right. Args: x (Tensor): The input Tensor with data type float32, float64. threshold (float, optional): The value of threshold for hardthrink. Default is 0.5. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-1, 0.3, 2.5]) out = F.hardshrink(x) # [-1., 0., 2.5] thresholdrr hardshrink hard_shrinkrrrN)r4) r rr5r rrr r#r$r%r&rr3r(r)r*r+r+r,r4s !   r4cCstr t|||Strt|d|d|St|dgddtd it}|j|j d}|j dd|id |i||d d |S)a hardtanh activation. Calculate the `hardtanh` of input `x`. .. math:: hardtanh(x)= \left\{ \begin{array}{cll} max,& & \text{if } x > max \\ min,& & \text{if } x < min \\ x,& & \text{otherwise} \end{array} \right. Parameters: x (Tensor): The input Tensor with data type float32, float64. min (float, optional): The minimum value of the linear region range. Default is -1. max (float, optional): The maximum value of the linear region range. Default is 1. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-1.5, 0.3, 2.5]) out = F.hardtanh(x) # [-1., 0.3, 1.] t_mint_maxrrhardtanhr%brelurr)r8r9rN)r:) r rr<r rrr r#r$r%r&)rminmaxr(r)r*r+r+r,r:s " r:UU?cCs~tr t|||Strt|d|d|St|dgddtd it}||j }|j dd|id|i||d d |S) a hardsigmoid activation. Calculate the `hardsigmoid` of input `x`. A 3-part piecewise linear approximation of sigmoid(https://arxiv.org/abs/1603.00391), which is much faster than sigmoid. .. math:: hardsigmoid(x)= \left\{ \begin{array}{lcl} 0, & &\text{if } \ x \leq -3 \\ 1, & &\text{if } \ x \geq 3 \\ slope * x + offset, & &\text{otherwise} \end{array} \right. Parameters: x (Tensor): The input Tensor with data type float32, float64. slope (float, optional): The slope of hardsigmoid function. Default is 0.1666667. offset (float, optional): The offset of hardsigmoid function. Default is 0.5. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-4., 5., 1.]) out = F.hardsigmoid(x) # [0., 1., 0.666667] slopeoffsetrr hardsigmoid hard_sigmoidrr)r@rArN)rB) r rrCr rrr r#r$r%r&)rr@rAr(r)r*r+r+r,rB=s $  rBcCsptrt|Strt|dddSt|dgddtd it}||j }|j dd|id|id |S) a hardswish activation. hardswish is proposed in MobileNetV3, and performs better in computational stability and efficiency compared to swish function. For more details please refer to: https://arxiv.org/pdf/1905.02244.pdf .. math:: hardswish(x)= \left\{ \begin{array}{cll} 0 &, & \text{if } x \leq -3 \\ x &, & \text{if } x \geq 3 \\ \frac{x(x+3)}{6} &, & \text{otherwise} \end{array} \right. Parameters: x (Tensor): The input Tensor with data type float32, float64. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-4., 5., 1.]) out = F.hardswish(x) # [0., 5., 0.666667] rrr hardswish hard_swishrrrrr N)rE) r rrFr rrr r#r$r%r&rr(r)r*r+r+r,rEvs"   rE{Gz?cCsxtr t||Strt|d|St|dgddtd it}|j|j d}|j dd|id|id|id|S) a leaky_relu activation. The calculation formula is: .. math:: leaky\_relu(x)= \left\{ \begin{array}{rcl} x, & & if \ x >= 0 \\ negative\_slope * x, & & otherwise \\ \end{array} \right. Args: x (Tensor): The input Tensor with data type float32, float64. negative_slope (float, optional): Slope of the activation function at :math:`x < 0` . Default is 0.01. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-2., 0., 1.]) out = F.leaky_relu(x) print(out) # [-0.02, 0., 1.] rrr leaky_relur;rrrN)rJ) r rrJr rrr r#r$r%r&)rZnegative_sloper(r)r*r+r+r,rJs "  rJNCHWcCsLt|dgddt|dgddt|jdksJdd}|jddkrmgd }||vr5td ||dd kr=d nd }t|jdksJJd|d kr]|jd|jdks\Jdn|jd|jdkskJdd}trxt||||Strt ||d|d|St dit }| |j }|jd||dd|i||dd|S)aY prelu activation. .. math:: prelu(x) = max(0, x) + weight * min(0, x) Parameters: x (Tensor): The input Tensor with data type float32, float64. weight (Tensor): The learnable parameter with data type same as ``x``. The weight shape is [1] or [in], where `in` is the input channel of ``x``. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. data_format(str, optional): Data format that specifies the layout of input. It may be "NC", "NCL", "NCHW", "NCDHW", "NLC", "NHWC" or "NDHWC". Default: "NCHW". Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F data = paddle.to_tensor([[[[-2.0, 3.0, -4.0, 5.0], [ 3.0, -4.0, 5.0, -6.0], [-7.0, -8.0, 8.0, 9.0]], [[ 1.0, -2.0, -3.0, 4.0], [-5.0, 6.0, 7.0, -8.0], [ 6.0, 7.0, 8.0, 9.0]]]], dtype='float32') w = paddle.to_tensor([0.25], dtype='float32') out = F.prelu(data, w) print(out) # [[[[-0.5 , 3. , -1. , 5. ], # [ 3. , -1. , 5. , -1.5 ], # [-1.75, -2. , 8. , 9. ]], # [[ 1. , -0.5 , -0.75, 4. ], # [-1.25, 6. , 7. , -2. ], # [ 6. , 7. , 8. , 9. ]]]] rrpreluweightz5The dim count of weight shape should be 1 in prelu().allr)ZNCNCLrKZNCDHWZNLCNHWCZNDHWCz^data_format must be one of 'NC', 'NCL', 'NCHW', 'NCDHW', 'NLC', 'NHWC', 'NDHWC' but receive {}CrKrQz\The dim count of x should be equal or larger than 2 in prelu() when weight shape is not [1].z[The weight size should be equal to x input channel in prelu() when weight shape is not [1].Zchannelmode data_format)rAlphar)rTrUrN)rL)rlenshape ValueErrorformatr rrLr rr r#r$r%r&)rrMrUr(rTZtrue_data_formatr)r*r+r+r,rLsT*    rL?UUUUUU?Tc Csts t|dgddt|trt|tstd|||dks&|dkr-td|||kr9td|||dkrDtd || }trYt |d |d |d |\}}|St dit }| |j }|j |j d }|||d} |jdd|i||d| d|S)a rrelu activation. Applies the randomized leaky rectified liner unit function to improve generalization performance, as described in the paper: `Empirical Evaluation of Rectified Activations in Convolutional Network `_ During training, randomly samples the negative slope for activation values as described below: .. math:: rrelu(x)= \left\{ \begin{array}{rcl} x, & & if \ x >= 0 \\ a * x, & & otherwise \\ \end{array} \right. where :math:`x` is the input tensor, :math:`a` is randomly sampled from uniform distribution in range (:math:`lower`, :math:`upper`), In the test phase, the negative slope will take the average value of :math:`lower` and :math:`upper`: .. math:: rrelu(x)= \left\{ \begin{array}{rcl} x, & & if \ x >= 0 \\ (lower + upper) * 0.5 * x, & & otherwise \\ \end{array} \right. where :math:`x` is the input tensor, :math:`lower` and :math:`upper` are the bounds of uniform distribution. Parameters: x (Tensor): The input Tensor with data type float16, float32, float64. lower (float, optional): The lower bound of uniform distribution. Default: 0.125. upper (float, optional): The upper bound of uniform distribution. Default: 0.333. training (bool, optional): Current mode is in training or others. Default is True. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F input_tensor = paddle.to_tensor([[[[-2.0, 3.0, -4.0, 5.0], [ 3.0, -4.0, 5.0, -6.0], [-7.0, -8.0, 8.0, 9.0]], [[ 1.0, -2.0, -3.0, 4.0], [-5.0, 6.0, 7.0, -8.0], [ 6.0, 7.0, 8.0, 9.0]]]], dtype='float32') out = F.rrelu(input_tensor, 0.1, 0.3) print(out) #[[[[-0.20000899 3. -0.8810822 5. ] # [ 3. -0.55175185 5. -1.0776101 ] # [-1.0680687 -1.9896201 8. 9. ]] # [[ 1. -0.5238267 -0.65515125 4. ] # [-1.3766339 6. 7. -2.3465784 ] # [ 6. 7. 8. 9. ]]]] rrrreluzLThe lower and upper values must be float type. Received: lower {}, upper {}.rrNzLThe lower value must be no less than zero or greater than one. Received: {}.zOThe upper value must be greater than lower value. Received: lower {}, upper {}.z:The upper value must be no greater than one. Received: {}.lowerupperis_testr;)r^r_r`)rZNoiserN)r])rr isinstancefloat TypeErrorrZrYr rr]r r#r$r%r&) rr^r_Ztrainingr(r`r*noiser)r!r+r+r,r]EsZG   r]cCjtrt|Strt|St|dgddtdit}||j }|j dd|id|id|S) al relu activation. .. math:: out = max(x, 0) Parameters: x (Tensor): The input Tensor with data type float32, float64. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-2, 0, 1], dtype='float32') out = F.relu(x) print(out) # [0., 0., 1.] rrrelurrrGN)rf) r rrfr rrr r#r$r%r&rHr+r+r,rfs   rfcCs$trt|Strt|SdS)z Inplace version of ``relu`` API, the output Tensor will be inplaced with input ``x``. Please refer to :ref:`api_nn_cn_relu`. N)r rrelu_r r)rr(r+r+r,rgs   rgcCjtrt|Strt|St|dgddtd it}||j }|j dd|id|id|S) a log_sigmoid activation. .. math:: log\_sigmoid(x) = log \frac{1}{1 + e^{-x}} Parameters: x (Tensor): The input Tensor with data type float32, float64. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([1.0, 2.0, 3.0, 4.0]) out = F.log_sigmoid(x) # [-0.313262 -0.126928 -0.0485874 -0.0181499] rr log_sigmoid logsigmoidrrrGN)ri) r rrjr rrr r#r$r%r&rHr+r+r,ris    rirNcCstr t|d|d|Strt|||St|dddgd|dvr+tdt||d kr1d }tdit }| |j }|j dd |id |i||d d|S)a maxout activation. Assumed the input shape is (N, Ci, H, W). The output shape is (N, Co, H, W). Then Co = Ci/groups and the operator formula is as follows: .. math:: \begin{array}{l} &out_{si+j} = \max_{k} x_{gsi + sk + j} \\ &g = groups \\ &s = \frac{input.size}{num\_channels} \\ &0 \le i < \frac{num\_channels}{groups} \\ &0 \le j < s \\ &0 \le k < groups \end{array} Parameters: x (Tensor): The input is 4-D Tensor with shape [N, C, H, W] or [N, H, W, C], the data type of input is float32 or float64. groups (int, optional): The groups number of maxout. `groups` specifies the index of channel dimension where maxout will be performed. This must be a factor of number of features. Default is 1. axis (int, optional): The axis along which to perform maxout calculations. It should be 1 when data format is NCHW, be -1 or 3 when data format is NHWC. If ``axis`` < 0, it works the same way as :math:`axis + D` , where D is the dimensions of ``x`` . ``axis`` only supports 1, 3 or -1. Default is 1. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.rand([1, 2, 3, 4]) # [[[[0.5002636 0.22272532 0.17402348 0.2874594 ] # [0.95313174 0.6228939 0.7129065 0.7087491 ] # [0.02879342 0.88725346 0.61093384 0.38833922]] # [[0.5231306 0.03807496 0.91661984 0.15602879] # [0.666127 0.616567 0.30741522 0.24044901] # [0.7142536 0.7351477 0.31588817 0.23782359]]]] out = F.maxout(x, groups=2) # [[[[0.5231306 0.22272532 0.91661984 0.2874594 ] # [0.95313174 0.6228939 0.7129065 0.7087491 ] # [0.7142536 0.88725346 0.61093384 0.38833922]]]] groupsaxisrrrmaxout)rNrSrzkAttr(axis) should be 1 when data format is NCHW, -1 or 3 when data format is NHWC. Received Attr(axis): %s.rSrrr)rkrlrN)rm) r rrmr rrrYstrr r#r$r%r&)rrkrlr(r)r*r+r+r,rms,6 rmcCszd}tr t||Strt|d|St|dgddtd it}||j }|j dd|id|id|id|S) ak relu6 activation .. math:: relu6(x) = min(max(0,x), 6) Parameters: x (Tensor): The input Tensor with data type float32, float64. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-1, 0.3, 6.5]) out = F.relu6(x) print(out) # [0, 0.3, 6] g@r3rrrelu6rrrN)ro) r rrorrrr r#r$r%r&)rr(r3r)r*r+r+r,rojs  ro2֫?,x?cCs|dkr td||dkrtd|tr t|||Str,t|d|d|St|dgdd tdit }| |j }|j d d |id |i||d d |S)aZ selu activation .. math:: selu(x)= scale * \left\{ \begin{array}{lcl} x,& &\text{if } \ x > 0 \\ alpha * e^{x} - alpha,& &\text{if } \ x <= 0 \end{array} \right. Parameters: x (Tensor): The input Tensor with data type float32, float64. scale (float, optional): The value of scale(must be greater than 1.0) for selu. Default is 1.0507009873554804934193349852946 alpha (float, optional): The value of alpha(must be no less than zero) for selu. Default is 1.6732632423543772848170429916717 name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([[0.0, 1.0],[2.0, 3.0]]) out = F.selu(x) print(out) # [[0, 1.050701],[2.101402, 3.152103]] rz1The scale must be greater than 1.0. Received: {}.rz2The alpha must be no less than zero. Received: {}.scalerrrselurr)rrrrN)rs) rYrZr rrsr rrr r#r$r%r&)rrrrr(r)r*r+r+r,rss,' rscCre) a silu activation .. math:: silu(x) = \frac{x}{1 + e^{-x}} Where :math:`x` is the input Tensor. Parameters: x (Tensor): The input Tensor with data type float32, float64. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as :attr:`x`. Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([1.0, 2.0, 3.0, 4.0]) out = F.silu(x) # [ 0.731059, 1.761594, 2.857722, 3.928055 ] rrsilurrrGN)rt) r rrtr rrr r#r$r%r&rHr+r+r,rts   rtrScCs,|durt|tjjst|}d}tr&|dur|nt||}t||St rB|dur/|n t |d|j d|}t |d|d|S|durPt |dgdd n t |d d d gd d tdit}|}|dur~||}|jdd|id|i|j |dd||j }|jd d|id|i||dd|S)a This operator implements the softmax layer. The calculation process is as follows: 1. The dimension :attr:`axis` of ``x`` will be permuted to the last. 2. Then ``x`` will be logically flattened to a 2-D matrix. The matrix's second dimension(row length) is the same as the dimension :attr:`axis` of ``x``, and the first dimension(column length) is the product of all other dimensions of ``x``. For each row of the matrix, the softmax operator squashes the K-dimensional(K is the width of the matrix, which is also the size of ``x``'s dimension :attr:`axis`) vector of arbitrary real values to a K-dimensional vector of real values in the range [0, 1] that add up to 1. 3. After the softmax operation is completed, the inverse operations of steps 1 and 2 are performed to restore the two-dimensional matrix to the same dimension as the ``x`` . It computes the exponential of the given dimension and the sum of exponential values of all the other dimensions in the K-dimensional vector input. Then the ratio of the exponential of the given dimension and the sum of exponential values of all the other dimensions is the output of the softmax operator. For each row :math:`i` and each column :math:`j` in the matrix, we have: .. math:: softmax[i, j] = \frac{\exp(x[i, j])}{\sum_j(exp(x[i, j])} Example: .. code-block:: text Case 1: Input: x.shape = [2, 3, 4] x.data = [[[2.0, 3.0, 4.0, 5.0], [3.0, 4.0, 5.0, 6.0], [7.0, 8.0, 8.0, 9.0]], [[1.0, 2.0, 3.0, 4.0], [5.0, 6.0, 7.0, 8.0], [6.0, 7.0, 8.0, 9.0]]] Attrs: axis = -1 Output: out.shape = [2, 3, 4] out.data = [[[0.0320586 , 0.08714432, 0.23688282, 0.64391426], [0.0320586 , 0.08714432, 0.23688282, 0.64391426], [0.07232949, 0.19661193, 0.19661193, 0.53444665]], [[0.0320586 , 0.08714432, 0.23688282, 0.64391426], [0.0320586 , 0.08714432, 0.23688282, 0.64391426], [0.0320586 , 0.08714432, 0.23688282, 0.64391426]]] Case 2: Input: x.shape = [2, 3, 4] x.data = [[[2.0, 3.0, 4.0, 5.0], [3.0, 4.0, 5.0, 6.0], [7.0, 8.0, 8.0, 9.0]], [[1.0, 2.0, 3.0, 4.0], [5.0, 6.0, 7.0, 8.0], [6.0, 7.0, 8.0, 9.0]]] Attrs: axis = 1 Output: out.shape = [2, 3, 4] out.data = [[[0.00657326, 0.00657326, 0.01714783, 0.01714783], [0.01786798, 0.01786798, 0.04661262, 0.04661262], [0.97555875, 0.97555875, 0.93623955, 0.93623955]], [[0.00490169, 0.00490169, 0.00490169, 0.00490169], [0.26762315, 0.26762315, 0.26762315, 0.26762315], [0.72747516, 0.72747516, 0.72747516, 0.72747516]]] Parameters: x (Tensor): The input Tensor with data type float32, float64. axis (int, optional): The axis along which to perform softmax calculations. It should be in range [-D, D), where D is the rank of ``x`` . If ``axis`` < 0, it works the same way as :math:`axis + D` . Default is -1. dtype (str, optional): The data type of the output tensor, can be float32, float64. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same shape and data type (use ``dtype`` if it is specified) as x. Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([[[2.0, 3.0, 4.0, 5.0], [3.0, 4.0, 5.0, 6.0], [7.0, 8.0, 8.0, 9.0]], [[1.0, 2.0, 3.0, 4.0], [5.0, 6.0, 7.0, 8.0], [6.0, 7.0, 8.0, 9.0]]],dtype='float32') out1 = F.softmax(x) out2 = F.softmax(x, dtype='float64') # out1's data type is float32; out2's data type is float64 # out1 and out2's value is as follows: # [[[0.0320586 , 0.08714432, 0.23688282, 0.64391426], # [0.0320586 , 0.08714432, 0.23688282, 0.64391426], # [0.07232949, 0.19661193, 0.19661193, 0.53444665]], # [[0.0320586 , 0.08714432, 0.23688282, 0.64391426], # [0.0320586 , 0.08714432, 0.23688282, 0.64391426], # [0.0320586 , 0.08714432, 0.23688282, 0.64391426]]] NTin_dtype out_dtyperl use_cudnnrrsoftmaxr%rr9If dtype is not None, it only support float32 or float64.castrrrurvr)rlrw)rx)rarVarDescVarTyper r rrzrxr rr%rrr r#r$r&)rrlr%r(rw outs_castr)Z outs_softmaxr+r+r,rxsVq      rxcCs|durt|tjjst|}d}tr*|dur|n t|d|jd|}t ||St rF|dur3|n t|d|jd|}t |d|d|SdS)z Inplace version of ``softmax`` API, the output Tensor will be inplaced with input ``x``. Please refer to :ref:`api_nn_cn_softmax`. NTrurvrlrw) rarr|r}r r rrzr%rsoftmax_r )rrlr%r(rwr~r+r+r,rs$  rcCs~tr t|||Strt|d|d|St|dgddtd it}||j }|j dd|id|i||dd |S) a softplus activation .. math:: softplus(x)=\begin{cases} \frac{1}{\beta} * \log(1 + e^{\beta * x}),&x\leqslant\frac{\varepsilon}{\beta};\\ x,&x>\frac{\varepsilon}{\beta}. \end{cases} Parameters: x (Tensor): The input Tensor with data type float32, float64. beta (float, optional): The value of :math:`\beta` for softplus. Default is 1 threshold (float, optional): The value of :math:`\varepsilon` for softplus. Default is 20 name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-0.4, -0.2, 0.1, 0.3], dtype='float32') out = F.softplus(x) # [0.513015, 0.598139, 0.744397, 0.854355] betar3rrsoftplusrr)rr3rN)r) r rrr rrr r#r$r%r&)rrr3r(r)r*r+r+r,rs   rcCs|dkr td|trt||Strt|d|St|dgddt d it }| |j }|j dd|id|id|id |S) au softshrink activation .. math:: softshrink(x)= \left\{ \begin{array}{rcl} x - threshold,& & \text{if } x > threshold \\ x + threshold,& & \text{if } x < -threshold \\ 0,& & \text{otherwise} \end{array} \right. Parameters: x (Tensor): The input Tensor with data type float32, float64. threshold (float, optional): The value of threshold(must be no less than zero) for softplus. Default is 0.5 name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-0.9, -0.2, 0.1, 0.8]) out = F.softshrink(x) print(out) # Tensor(shape=[4], dtype=float32, place=Place(gpu:0), stop_gradient=True, # [-0.39999998, 0. , 0. , 0.30000001]) rz6The threshold must be no less than zero. Received: {}.lambdarr softshrinkrrrN)r)rYrZr rZ soft_shrinkr rrrr r#r$r%r&r6r+r+r,rs,#   rcCre) a softsign activation .. math:: softsign(x) = \frac{x}{1 + |x|} Parameters: x (Tensor): The input Tensor with data type float32, float64. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-0.4, -0.2, 0.1, 0.3]) out = F.softsign(x) print(out) # Tensor(shape=[4], dtype=float32, place=Place(gpu:0), stop_gradient=True, # [-0.28571430, -0.16666666, 0.09090909, 0.23076925]) rrsoftsignrrrGN)r) r rrrrrr r#r$r%r&rHr+r+r,r0s    rcCsvtr t|dStrt|ddSt|dgddtd it}||j }|j dd|id|iddid|S) a swish activation. .. math:: swish(x) = \frac{x}{1 + e^{-x}} Parameters: x (Tensor): The input Tensor with data type float32, float64. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-2., 0., 1.]) out = F.swish(x) print(out) # Tensor(shape=[3], dtype=float32, place=Place(gpu:0), stop_gradient=True, # [-0.23840584, 0. , 0.73105854]) rrrrswishrrrN)r) r rrr rrr r#r$r%r&rHr+r+r,rYs  rcCsltr t|dStrt|St|dgddtd it}||j }|j dd|id|id|S) a mish activation. .. math:: softplus(x) = \begin{cases} x, \text{if } x > \text{threshold} \\ \ln(1 + e^{x}), \text{otherwise} \end{cases} mish(x) = x * \tanh(softplus(x)) Parameters: x (Tensor): The input Tensor with data type float32, float64. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-5., 0., 5.]) out = F.mish(x) # [-0.03357624, 0., 4.99955208] rrrmishrrrGN)r) r rrr rrr r#r$r%r&rHr+r+r,rs   rcCrh) a tanhshrink activation .. math:: tanhshrink(x) = x - tanh(x) Args: x (Tensor): The input Tensor with data type float32, float64. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([-0.4, -0.2, 0.1, 0.3]) out = F.tanhshrink(x) print(out) # Tensor(shape=[4], dtype=float32, place=Place(gpu:0), stop_gradient=True, # [-0.02005106, -0.00262468, 0.00033200, 0.00868741]) rr tanhshrink tanh_shrinkrrrGN)r) r rrr rrr r#r$r%r&rHr+r+r,rs    rcCr-) a thresholded relu activation. .. math:: thresholded\_relu(x) = \left\{ \begin{array}{rl} x,& \text{if } \ x > threshold \\ 0,& \text{otherwise} \end{array} \right. Parameters: x (Tensor): The input Tensor with data type float32, float64. threshold (float, optional): The value of threshold for thresholded_relu. Default is 1.0 name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type and shape as ``x`` . Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = paddle.to_tensor([2., 0., 1.]) out = F.thresholded_relu(x) print(out) # Tensor(shape=[3], dtype=float32, place=Place(gpu:0), stop_gradient=True, # [2., 0., 0.]) r3rrthresholded_relurrrN)r) r rrr rrr r#r$r%r&r6r+r+r,rs $   rcCs|durt|tjjst|}tr"|durt||}t||St r:|dur3t |d|j d|}t |d|S|durHt |dgddn t |dd d gdd tdit}|}|durv||}|jd d |id|i|j |dd||j }|jdd |id|id|id|S)a This operator implements the log_softmax layer. The calculation process is as follows: .. math:: \begin{aligned} log\_softmax[i, j] &= log(softmax(x)) \\ &= log(\frac{\exp(X[i, j])}{\sum_j(\exp(X[i, j])}) \end{aligned} Parameters: x (Tensor): The input Tensor with data type float32, float64. axis (int, optional): The axis along which to perform log_softmax calculations. It should be in range [-D, D), where D is the dimensions of ``x`` . If ``axis`` < 0, it works the same way as :math:`axis + D` . Default is -1. dtype (str|np.dtype|core.VarDesc.VarType, optional): The desired data type of the output tensor. If dtype is specified, ``x`` is casted to ``dtype`` before the operation is performed. This is useful for preventing data type overflows. Supported dtype: float32, float64. If ``dtype`` is None, the output Tensor has the same dtype as x. Default is None. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same shape and data type (use ``dtype`` if it is specified) as x. Examples: .. code-block:: python import paddle import paddle.nn.functional as F x = [[[-2.0, 3.0, -4.0, 5.0], [3.0, -4.0, 5.0, -6.0], [-7.0, -8.0, 8.0, 9.0]], [[1.0, -2.0, -3.0, 4.0], [-5.0, 6.0, 7.0, -8.0], [6.0, 7.0, 8.0, 9.0]]] x = paddle.to_tensor(x) out1 = F.log_softmax(x) out2 = F.log_softmax(x, dtype='float64') # out1's data type is float32; out2's data type is float64 # out1 and out2's value is as follows: # [[[ -7.1278396 -2.1278396 -9.127839 -0.12783948] # [ -2.1270514 -9.127051 -0.12705144 -11.127051 ] # [-16.313261 -17.313261 -1.3132617 -0.31326184]] # [[ -3.0518122 -6.051812 -7.051812 -0.051812 ] # [-12.313267 -1.3132664 -0.3132665 -15.313267 ] # [ -3.4401896 -2.4401896 -1.4401896 -0.44018966]]] Nrurvrlrr log_softmaxr%rrryrzrrr{r)r)rarr|r}r r rrzrr rr%rrr r#r$r&)rrlr%r(r)Zout_castr*r+r+r,r sN7      rcCsFt|dgddt|d||d\}}t||d}tj|||d}|S)a The gated linear unit. The input is evenly splited into 2 parts along a given axis. The first part is used as the content, and the second part is passed through a sigmoid function then used as the gate. The output is a elementwise multiplication of the content and the gate. .. math:: \mathrm{GLU}(a, b) = a \otimes \sigma(b) Parameters: x (Tensor): The input Tensor with data type float32, float64. axis (int, optional): The axis along which split the input tensor. It should be in range [-D, D), where D is the dimensions of ``x`` . If ``axis`` < 0, it works the same way as :math:`axis + D` . Default is -1. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: A Tensor with the same data type as x. The size of the given aixs is halved. Examples: .. code-block:: python import paddle from paddle.nn import functional as F x = paddle.to_tensor( [[-0.22014759, -1.76358426, 0.80566144, 0.04241343], [-1.94900405, -1.89956081, 0.17134808, -1.11280477]] ) print(F.glu(x).numpy()) # array([[-0.15216254, -0.9004892 ], # [-1.0577879 , -0.46985325]], dtype=float32) inputrglu)rlr()r()rrrpaddler)rrlr(abZgater*r+r+r,rts&  rc Cstr t||||Strt|d|d|d|Std it}t|dddgd||j }|j dd|id |i|||d d |S)a Samples from the Gumbel-Softmax distribution and optionally discretizes. temperature is denoted by t. The calculation process is as follows: First, generate gumbel noise: .. math:: G_i = -log(-log(U_i)), U_i \sim U(0,1) Second, add noise to ``x``: .. math:: v = [x_1 + G_1,...,x_n + G_n] Finally, calculate gumbel_softmax and generate samples: .. math:: gumbel\_softmax(v_i)=\frac{e^{v_i/t}}{\sum_{j=1}^n{e^{v_j/t}}},i=1,2,3...n Parameters: x (Tensor): An N-D Tensor, the first N - 1 dimensions index into a batch of independent distributions and the last dimension represents a vector of probabilities with datatype float32, float64. temperature (float, optional): non-negative scalar temperature. Default is 1.0. hard (bool, optional): if True, the returned samples will be discretized as one-hot vectors, but will be differentiated as if it is the soft sample in autograd. Default is False. axis (int, optional): The axis along will be calculated softmax value. Default is -1. name (str, optional): For details, please refer to :ref:`api_guide_Name`. Generally, no setting is required. Default: None. Returns: Sampled tensor of same shape as ``x`` from the Gumbel-Softmax distribution. If ``hard = True``, the returned samples will be one-hot, otherwise they will be probability distributions that sum to 1 across ``axis``. Examples: .. code-block:: python import paddle import paddle.nn.functional as F logits = paddle.randn([4, 6]) temperature = 0.01 gumbel_softmax = F.gumbel_softmax(logits, temperature) print(gumbel_softmax) # out's value is as follows: # [[0.00000001, 1. , 0.00000000, 0.00000000, 0.00000006, 0.00000000], # [0.00000000, 0.00000000, 0.00000000, 0.00000000, 0.00000000, 1. ], # [0.00000062, 0.00000000, 0.00000000, 0.00000000, 0.00000000, 0.99999940], # [0.00000000, 0.00000000, 0.00000000, 0.00001258, 0.99998736, 0.00000000]] temperaturehardrlgumbel_softmaxrrrrr)rrrlrN)r) r rrrrr r#rr$r%r&)rrrrlr(r)r*r+r+r,rs 9  r)rN)FN)r2N)r7rN)r?r2N)N)rIN)rKN)r[r\TN)rNN)rprqN)rSNN)rNrN)rSN)rFrSN);Z tensor.opsrZ tensor.mathrrZfluid.dygraph.inplace_utilsrZtensor.manipulationrrwarningsZfluid.layer_helperr Zfluid.frameworkr r r r Zfluid.data_feederrrrrrrZpaddle.frameworkrZpaddle.fluid.framework__all__rr.r/r1r4r:rBrErJrLr]rfrgrirmrorsrtrxrrrrrrrrrrrr+r+r+r,sj           .4  ; 5 7 9 1 6 h }&  ( N. B ''   1 = ) ) ) * 8 g/