o Me/@sddlZddlZddlZddlZddlmZmZddlmZddl m Z ddl m Z m Z mZmZddlmZmZddlmZmZmZmZmZmZmZmZddlmZmZmZm Z Gd d d ej!Z"dS) N)_C_ops _legacy_C_ops) distribution)core) check_dtype check_typecheck_variable_and_dtype convert_dtype)_non_static_modein_dygraph_mode) control_flowelementwise_addelementwise_divelementwise_mulelementwise_subnnopstensor)arangeconcat gather_nd multinomialc@sBeZdZdZdddZddZddZd d Zd d Zd dZ dS) Categoricala Categorical distribution is a discrete probability distribution that describes the possible results of a random variable that can take on one of K possible categories, with the probability of each category separately specified. The probability mass function (pmf) is: .. math:: pmf(k; p_i) = \prod_{i=1}^{k} p_i^{[x=i]} In the above equation: * :math:`[x=i]` : it evaluates to 1 if :math:`x==i` , 0 otherwise. Args: logits(list|tuple|numpy.ndarray|Tensor): The logits input of categorical distribution. The data type is float32 or float64. name(str, optional): Name for the operation (optional, default is None). For more information, please refer to :ref:`api_guide_Name`. Examples: .. code-block:: python import paddle from paddle.distribution import Categorical paddle.seed(100) # on CPU device x = paddle.rand([6]) print(x) # [0.5535528 0.20714243 0.01162981 # 0.51577556 0.36369765 0.2609165 ] paddle.seed(200) # on CPU device y = paddle.rand([6]) print(y) # [0.77663314 0.90824795 0.15685187 # 0.04279523 0.34468332 0.7955718 ] cat = Categorical(x) cat2 = Categorical(y) paddle.seed(1000) # on CPU device cat.sample([2,3]) # [[0, 0, 5], # [3, 4, 5]] cat.entropy() # [1.77528] cat.kl_divergence(cat2) # [0.071952] value = paddle.to_tensor([2,1,3]) cat.probs(value) # [0.00608027 0.108298 0.269656] cat.log_prob(value) # [-5.10271 -2.22287 -1.31061] NcCstst|dtjtjttfd|dur|nd|_d|_ | |r+||_ t |j |_ n,t |tjr||dd}t |jt|dd|dg}n|}|j}t | ||d}tt|}|d|d||}t j |||d S) aGenerate samples of the specified shape. Args: shape (list): Shape of the generated samples. Returns: Tensor: A tensor with prepended dimensions shape. Examples: .. code-block:: python import paddle from paddle.distribution import Categorical paddle.seed(100) # on CPU device x = paddle.rand([6]) print(x) # [0.5535528 0.20714243 0.01162981 # 0.51577556 0.36369765 0.2609165 ] cat = Categorical(x) paddle.seed(1000) # on CPU device cat.sample([2,3]) # [[0, 0, 5], # [3, 4, 5]] Z_sampleshapesampleNrTrr$)r$r rr"r prodarrayrr/lenr(reshaperZ_logits_to_probsrangediminsertpopZ transpose) r+r/r$Z num_samplesZ logits_shapeZ sample_shaperZ sample_indexZpermuter,r,r-r0ys&     zCategorical.samplec Cs|jd}tst|dtd|jtj|jddd}|jtj|jddd}t|}t|}tj |ddd}tj |ddd}||} tj | |t ||t |dd|d} | S)aThe KL-divergence between two Categorical distributions. Args: other (Categorical): instance of Categorical. The data type is float32. Returns: Tensor: kl-divergence between two Categorical distributions. Examples: .. code-block:: python import paddle from paddle.distribution import Categorical paddle.seed(100) # on CPU device x = paddle.rand([6]) print(x) # [0.5535528 0.20714243 0.01162981 # 0.51577556 0.36369765 0.2609165 ] paddle.seed(200) # on CPU device y = paddle.rand([6]) print(y) # [0.77663314 0.90824795 0.15685187 # 0.04279523 0.34468332 0.7955718 ] cat = Categorical(x) cat2 = Categorical(y) cat.kl_divergence(cat2) # [0.071952] Z_kl_divergenceother kl_divergencerTr)rrr$) r$r rrrr(maxrexpr)log) r+r;r$rZ other_logitse_logitsZother_e_logitszZother_zprobklr,r,r-r<s. "  zCategorical.kl_divergencecCst|jd}|jtj|jddd}t|}tj|ddd}||}tj||t|dd}tj|d|d}|S)aZShannon entropy in nats. Returns: Tensor: Shannon entropy of Categorical distribution. The data type is float32. Examples: .. code-block:: python import paddle from paddle.distribution import Categorical paddle.seed(100) # on CPU device x = paddle.rand([6]) print(x) # [0.5535528 0.20714243 0.01162981 # 0.51577556 0.36369765 0.2609165 ] cat = Categorical(x) cat.entropy() # [1.77528] Z_entropyrTrrg)scaler$) r$rr(r=rr>r)r?rE)r+r$rr@rArBZ neg_entropyentropyr,r,r-rFs  zCategorical.entropycCs|jd}t|jjdkr"tj|j|jdg|d|dj|j|dSt|jdkrDtj|jtj|t|jjddgdg|dddStj|j|ddS)aProbabilities of the given category (``value``). If ``logits`` is 2-D or higher dimension, the last dimension will be regarded as category, and the others represents the different distributions. At the same time, if ``vlaue`` is 1-D Tensor, ``value`` will be broadcast to the same number of distributions as ``logits``. If ``value`` is not 1-D Tensor, ``value`` should have the same number distributions with ``logits. That is, ``value[:-1] = logits[:-1]``. Args: value (Tensor): The input tensor represents the selected category index. Returns: Tensor: probability according to the category index. Examples: .. code-block:: python import paddle from paddle.distribution import Categorical paddle.seed(100) # on CPU device x = paddle.rand([6]) print(x) # [0.5535528 0.20714243 0.01162981 # 0.51577556 0.36369765 0.2609165 ] cat = Categorical(x) value = paddle.to_tensor([2,1,3]) cat.probs(value) # [0.00608027 0.108298 0.269656] Z_probsr1rr2rD)r$r5r*r/r(gatherr6Ztake_along_axisr+valuer$r,r,r-probs s$ #zCategorical.probscCs|jd}tj|||dS)aLog probabilities of the given category. Refer to ``probs`` method. Args: value (Tensor): The input tensor represents the selected category index. Returns: Tensor: Log probability. Examples: .. code-block:: python import paddle from paddle.distribution import Categorical paddle.seed(100) # on CPU device x = paddle.rand([6]) print(x) # [0.5535528 0.20714243 0.01162981 # 0.51577556 0.36369765 0.2609165 ] cat = Categorical(x) value = paddle.to_tensor([2,1,3]) cat.log_prob(value) # [-5.10271 -2.22287 -1.31061] Z _log_probr2)r$r(r?rJrHr,r,r-log_prob>s zCategorical.log_prob)N) __name__ __module__ __qualname____doc__r.r0r<rFrJrKr,r,r,r-r s =78# 3r)#mathwarningsnumpyr r(rrZpaddle.distributionrZ paddle.fluidrZpaddle.fluid.data_feederrrrr Zpaddle.fluid.frameworkr r Zpaddle.fluid.layersr r rrrrrrZ paddle.tensorrrrr Distributionrr,r,r,r-s  (