U
    d                     @   sT   d dl Z d dlmZ d dlmZ d dlmZ d dlmZm	Z	m
Z
 G dd deZdS )    N)nan)constraints)Distribution)probs_to_logitslogits_to_probslazy_propertyc                       s   e Zd ZdZejejdZdZd# fdd	Z	d$ fdd	Z
d	d
 Zejddddd Zedd Zedd Zedd Zedd Zedd Zedd Ze fddZdd Zdd  Zd%d!d"Z  ZS )&Categoricala  
    Creates a categorical distribution parameterized by either :attr:`probs` or
    :attr:`logits` (but not both).

    .. note::
        It is equivalent to the distribution that :func:`torch.multinomial`
        samples from.

    Samples are integers from :math:`\{0, \ldots, K-1\}` where `K` is ``probs.size(-1)``.

    If `probs` is 1-dimensional with length-`K`, each element is the relative probability
    of sampling the class at that index.

    If `probs` is N-dimensional, the first N-1 dimensions are treated as a batch of
    relative probability vectors.

    .. note:: The `probs` argument must be non-negative, finite and have a non-zero sum,
              and it will be normalized to sum to 1 along the last dimension. :attr:`probs`
              will return this normalized value.
              The `logits` argument will be interpreted as unnormalized log probabilities
              and can therefore be any real number. It will likewise be normalized so that
              the resulting probabilities sum to 1 along the last dimension. :attr:`logits`
              will return this normalized value.

    See also: :func:`torch.multinomial`

    Example::

        >>> m = Categorical(torch.tensor([ 0.25, 0.25, 0.25, 0.25 ]))
        >>> m.sample()  # equal probability of 0, 1, 2, 3
        tensor(3)

    Args:
        probs (Tensor): event probabilities
        logits (Tensor): event log probabilities (unnormalized)
    )probslogitsTNc                    s   |d k|d kkrt d|d k	rJ| dk r4t d||jddd | _n(| dk r^t d||jddd | _|d k	r| jn| j| _| j d | _| j	 dkr| j d d nt
 }tt| j||d	 d S )
Nz;Either `probs` or `logits` must be specified, but not both.   z3`probs` parameter must be at least one-dimensional.T)keepdimz4`logits` parameter must be at least one-dimensional.)dimr   validate_args)
ValueErrorr   sumr	   Z	logsumexpr
   _paramsize_num_eventsZ
ndimensiontorchSizesuperr   __init__)selfr	   r
   r   batch_shape	__class__ C/tmp/pip-unpacked-wheel-ua33x9lu/torch/distributions/categorical.pyr   1   s    (zCategorical.__init__c                    s   |  t|}t|}|t| jf }d| jkrH| j||_|j|_d| jkrh| j	||_	|j	|_| j|_t
t|j|dd | j|_|S )Nr	   r
   Fr   )Z_get_checked_instancer   r   r   r   __dict__r	   expandr   r
   r   r   _validate_args)r   r   Z	_instancenewparam_shaper   r   r   r!   B   s    


zCategorical.expandc                 O   s   | j j||S N)r   r#   )r   argskwargsr   r   r   _newQ   s    zCategorical._newr   )Zis_discreteZ	event_dimc                 C   s   t d| jd S )Nr   r   )r   Zinteger_intervalr   r   r   r   r   supportT   s    zCategorical.supportc                 C   s
   t | jS r%   )r   r	   r)   r   r   r   r
   X   s    zCategorical.logitsc                 C   s
   t | jS r%   )r   r
   r)   r   r   r   r	   \   s    zCategorical.probsc                 C   s
   | j  S r%   )r   r   r)   r   r   r   r$   `   s    zCategorical.param_shapec                 C   s   t j|  t| jj| jjdS Ndtypedevicer   full_extended_shaper   r	   r-   r.   r)   r   r   r   meand   s    zCategorical.meanc                 C   s   | j jddS )Nr   )Zaxis)r	   Zargmaxr)   r   r   r   modeh   s    zCategorical.modec                 C   s   t j|  t| jj| jjdS r+   r/   r)   r   r   r   variancel   s    zCategorical.variancec                 C   sJ   t |tjst|}| jd| j}t|| dj}|| 	|S )Nr   T)

isinstancer   r   r	   Zreshaper   ZmultinomialZnumelTr1   )r   Zsample_shapeZprobs_2dZ
samples_2dr   r   r   samplep   s
    
zCategorical.samplec                 C   sR   | j r| | | d}t|| j\}}|dd df }|d|dS )Nr   .r   )	r"   Z_validate_samplelongZ	unsqueezer   Zbroadcast_tensorsr
   ZgatherZsqueeze)r   valueZlog_pmfr   r   r   log_probw   s    
zCategorical.log_probc                 C   s6   t | jjj}t j| j|d}|| j }|d S )N)minr   )r   Zfinfor
   r-   r;   clampr	   r   )r   Zmin_realr
   Zp_log_pr   r   r   entropy   s    
zCategorical.entropyc                 C   sL   | j }tj|tj| jjd}|ddt| j  }|rH|	d| j }|S )Nr,   )r   )r   )
r   r   Zaranger8   r   r.   viewlenZ_batch_shaper!   )r   r!   Z
num_eventsvaluesr   r   r   enumerate_support   s    zCategorical.enumerate_support)NNN)N)T)__name__
__module____qualname____doc__r   ZsimplexZreal_vectorZarg_constraintsZhas_enumerate_supportr   r!   r(   Zdependent_propertyr*   r   r
   r	   propertyr$   r2   r3   r4   r   r   r7   r:   r=   rA   __classcell__r   r   r   r   r      s4   $






r   )r   Z
torch._sixr   Ztorch.distributionsr   Z torch.distributions.distributionr   Ztorch.distributions.utilsr   r   r   r   r   r   r   r   <module>   s
   