o * iO7@sUddlmZddlmZmZddlZddlZddlm Z m Z ddl m Z ddl mZddlmZddlmZervdd lmZddlmZdd lmZdd lmZdd lmZdd lmZeeeeeejeej ej!fefZ"de#d<Gdddej$Z%dS)) annotations) TYPE_CHECKINGUnionN) check_type convert_dtype)Variable) distribution)in_dynamic_mode) multinomial)Sequence) TypeAlias)Tensor)NestedSequence) _DTypeLiteralr _CategoricalBoundaryc@sfeZdZUdZded<ded< ddd d ZgfdddZdddZd ddZd!ddZ d!ddZ 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|None, 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) Tensor(shape=[6], dtype=float32, place=Place(cpu), stop_gradient=True, [0.55355281, 0.20714243, 0.01162981, 0.51577556, 0.36369765, 0.26091650]) >>> paddle.seed(200) # on CPU device >>> y = paddle.rand([6]) >>> print(y) Tensor(shape=[6], dtype=float32, place=Place(cpu), stop_gradient=True, [0.77663314, 0.90824795, 0.15685187, 0.04279523, 0.34468332, 0.79557180]) >>> cat = Categorical(x) >>> cat2 = Categorical(y) >>> paddle.seed(1000) # on CPU device >>> print(cat.sample([2, 3])) Tensor(shape=[2, 3], dtype=int64, place=Place(cpu), stop_gradient=True, [[0, 1, 5], [3, 4, 5]]) >>> print(cat.entropy()) Tensor(shape=[], dtype=float32, place=Place(cpu), stop_gradient=True, 1.77528250) >>> print(cat.kl_divergence(cat2)) Tensor(shape=[1], dtype=float32, place=Place(cpu), stop_gradient=True, [0.07195196]) >>> value = paddle.to_tensor([2, 1, 3]) >>> print(cat.probs(value)) Tensor(shape=[3], dtype=float32, place=Place(cpu), stop_gradient=True, [0.00608027, 0.10829761, 0.26965630]) >>> print(cat.log_prob(value)) Tensor(shape=[3], dtype=float32, place=Place(cpu), stop_gradient=True, [-5.10270691, -2.22287226, -1.31060708]) r logitsrdtypeNrname str | NonereturnNonecCstst|dtjttjjtt fd|dur|nd|_ d|_ | |r-||_ t|j |_ n.t|tjr@t|j dvr@t|j |_ ||d|_ |j t|j j kr[tj|j |j d|_ tj|j dd d }|j ||_dS) a1 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`. rrNfloat32)rfloat64r)rTaxiskeepdim)r rnpZndarrayrpaddleZpirValuelisttuplerrZ_validate_argsrr isinstancestrZ _to_tensorcastsum_prob)selfrrZdist_sumr)k/home/app/PaddleOCR-VL-test/.venv_paddleocr/lib/python3.10/site-packages/paddle/distribution/categorical.py__init__ss&   zCategorical.__init__shape Sequence[int]c Cs|jd}tst|dttfdtt|}t|jj }t |dkr@||dd}t |jt|dd|dg}n|}|j}t |||d}tt|}|d|d||}t j |||d S) aGenerate samples of the specified shape. Args: shape (Sequence[int], optional): 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) Tensor(shape=[6], dtype=float32, place=Place(cpu), stop_gradient=True, [0.55355281, 0.20714243, 0.01162981, 0.51577556, 0.36369765, 0.26091650]) >>> # doctest: +SKIP('Random output') >>> cat = Categorical(x) >>> paddle.seed(1000) # on CPU device >>> print(cat.sample([2, 3])) Tensor(shape=[2, 3], dtype=int64, place=Place(cpu), stop_gradient=True, [[0, 1, 5], [3, 4, 5]]) Z_sampler,sampleNrTrr)rr rr!r"rprodarrayrr,lenrreshaper Z_logits_to_probsrangediminsertpopZ transpose) r(r,rZ num_samplesZ logits_shapeZ sample_shaperZ sample_indexZpermuter)r)r*r.s&     zCategorical.sampleotherc 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) Tensor(shape=[6], dtype=float32, place=Place(cpu), stop_gradient=True, [0.55355281, 0.20714243, 0.01162981, 0.51577556, 0.36369765, 0.26091650]) >>> paddle.seed(200) # on CPU device >>> y = paddle.rand([6]) >>> print(y) Tensor(shape=[6], dtype=float32, place=Place(cpu), stop_gradient=True, [0.77663314, 0.90824795, 0.15685187, 0.04279523, 0.34468332, 0.79557180]) >>> cat = Categorical(x) >>> cat2 = Categorical(y) >>> print(cat.kl_divergence(cat2)) Tensor(shape=[1], dtype=float32, place=Place(cpu), stop_gradient=True, [0.07195196]) Z_kl_divergencer9 kl_divergencerTr)rrr) rr rrrrmaxexpr&log) r(r9rrZ 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)aShannon 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) Tensor(shape=[6], dtype=float32, place=Place(cpu), stop_gradient=True, [0.55355281, 0.20714243, 0.01162981, 0.51577556, 0.36369765, 0.26091650]) >>> cat = Categorical(x) >>> print(cat.entropy()) Tensor(shape=[], dtype=float32, place=Place(cpu), stop_gradient=True, 1.77528250) Z_entropyrTrrg)scaler)rrrr;r<r&r=rC)r(rrr>r?r@Z neg_entropyentropyr)r)r*rD s  zCategorical.entropyvaluecCs|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 ``value`` 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) Tensor(shape=[6], dtype=float32, place=Place(cpu), stop_gradient=True, [0.55355281, 0.20714243, 0.01162981, 0.51577556, 0.36369765, 0.26091650]) >>> cat = Categorical(x) >>> value = paddle.to_tensor([2, 1, 3]) >>> print(cat.probs(value)) Tensor(shape=[3], dtype=float32, place=Place(cpu), stop_gradient=True, [0.00608027, 0.10829761, 0.26965630]) Z_probsr/rr0rB)rr3r'r,rgatherr4Ztake_along_axisr(rErr)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) Tensor(shape=[6], dtype=float32, place=Place(cpu), stop_gradient=True, [0.55355281, 0.20714243, 0.01162981, 0.51577556, 0.36369765, 0.26091650]) >>> cat = Categorical(x) >>> value = paddle.to_tensor([2, 1, 3]) >>> print(cat.log_prob(value)) Tensor(shape=[3], dtype=float32, place=Place(cpu), stop_gradient=True, [-5.10270691, -2.22287226, -1.31060708]) Z _log_probr0)rrr=rHrGr)r)r*log_probas zCategorical.log_prob)N)rrrrrr)r,r-rr )r9rrr )rr )rEr rr ) __name__ __module__ __qualname____doc____annotations__r+r.r:rDrHrIr)r)r)r*r.s A $ 9 9 "6r)& __future__rtypingrrnumpyrrZpaddle.base.data_feederrrZpaddle.base.frameworkrZpaddle.distributionrZpaddle.frameworkr Z paddle.tensorr collections.abcr Z numpy.typingZnptZtyping_extensionsr r Zpaddle._typingrZpaddle._typing.dtype_likerfloatZNDArrayrrrrN Distributionrr)r)r)r*s0