U
    8‹ÑhüA  ã                   @   s`  d dl mZmZmZ d dlZd dlmZ ddlmZmZm	Z	m
Z
mZmZmZmZmZmZmZmZ ddgZG dd„ deƒZd	d
e› de› de› de	› d�	 e_ee ee ee ee ee eeeeeeeedœdd„Zee ee ee ee ee eeeeeeeedœdd„Ze
ed�dee ee ee ee ee ee eeeeeeeedœdd„ƒZdS )é    )ÚListÚOptionalÚTupleN)ÚTensoré   )Ú_capturable_docÚ_default_to_fused_or_foreachÚ_differentiable_docÚ_disable_dynamo_if_unsupportedÚ_foreach_docÚ!_get_capturable_supported_devicesÚ_get_scalar_dtypeÚ_maximize_docÚ_use_grad_for_differentiableÚ_view_as_realÚ	OptimizerÚParamsTÚRpropÚrpropc                       st   e Zd Zddddddœeeeeef eeef eee eedœ‡ fdd	„Z‡ fd
d„Z	dd„ Z
eddd„ƒZ‡  ZS )r   ç{®Gáz„?©g      à?g333333ó?©g�íµ ÷Æ°>é2   FN)Ú
capturableÚforeachÚmaximizeÚdifferentiable)ÚparamsÚlrÚetasÚ
step_sizesr   r   r   r   c          
   	      s‚   d|kst d|› �ƒ‚d|d   k r<d  k r<|d k sZn t d|d › d|d › �ƒ‚t|||||||d�}	tƒ  ||	¡ d S )	Ng        zInvalid learning rate: r   ç      ð?r   zInvalid eta values: z, )r   r   r    r   r   r   r   )Ú
ValueErrorÚdictÚsuperÚ__init__)
Úselfr   r   r   r    r   r   r   r   Údefaults©Ú	__class__© ú?/tmp/user/2202/pip-unpacked-wheel-bf9o3bgk/torch/optim/rprop.pyr%      s    (ù	zRprop.__init__c                    s¾   t ƒ  |¡ | jD ]¦}| dd ¡ | dd¡ | dd¡ | dd¡ |d D ]h}| j |g ¡}t|ƒdkrNt |d ¡sNt	|d ƒ}|d r¢tj
|tƒ |jd	�ntj
|tƒ d
�|d< qNqd S )Nr   r   Fr   r   r   r   Ústep©ÚdtypeÚdevice©r.   )r$   Ú__setstate__Úparam_groupsÚ
setdefaultÚstateÚgetÚlenÚtorchZ	is_tensorÚfloatÚtensorr   r/   )r&   r4   ÚgroupÚpZp_stateZstep_valr(   r*   r+   r1   5   s$    
ý  ÿûzRprop.__setstate__c                 C   s  d}|d D �]}|j d krq|t |¡O }| |¡ |j }	|	jrJtdƒ‚| |	¡ | j| }
t|
ƒdkrê|d r†tjdt	ƒ |j
d�ntjdt	ƒ d�|
d	< tj|tjd
�|
d< |jjrÖt |	t|d |d ƒ¡|
d< nt |	|d ¡|
d< | |
d ¡ | |
d ¡ | |
d	 ¡ q|S )NFr   z'Rprop does not support sparse gradientsr   r   r*   r-   r0   r,   ©Zmemory_formatÚprevr   Ú	step_size)Úgradr7   Ú
is_complexÚappendZ	is_sparseÚRuntimeErrorr4   r6   Úzerosr   r/   Z
zeros_likeÚpreserve_formatr.   Z	full_likeÚcomplex)r&   r:   r   ÚgradsÚprevsr    Ústate_stepsÚhas_complexr;   r?   r4   r*   r*   r+   Ú_init_groupH   s6    



ÿý ÿ
zRprop._init_groupc                 C   sº   |   ¡  d}|dk	r.t ¡ � |ƒ }W 5 Q R X | jD ]€}g }g }g }g }g }|d \}	}
|d \}}|d }|d }|  ||||||¡}t||||||||	|
|||d |d |d� q4|S )	z±Performs a single optimization step.

        Args:
            closure (Callable, optional): A closure that reevaluates the model
                and returns the loss.
        Nr   r    r   r   r   r   )	Ústep_size_minÚstep_size_maxÚetaminusÚetaplusr   r   r   r   rI   )Z _cuda_graph_capture_health_checkr7   Zenable_gradr2   rJ   r   )r&   ÚclosureZlossr:   r   rF   rG   r    rH   rM   rN   rK   rL   r   r   rI   r*   r*   r+   r,   n   sP    

     ÿòz
Rprop.step)r   r   r   )N)Ú__name__Ú
__module__Ú__qualname__r   r8   r   Úboolr   r%   r1   rJ   r   r,   Ú__classcell__r*   r*   r(   r+   r      s,      ûö

ö&aÁ
  Implements the resilient backpropagation algorithm.

    .. math::
       \begin{aligned}
            &\rule{110mm}{0.4pt}                                                                 \\
            &\textbf{input}      : \theta_0 \in \mathbf{R}^d \text{ (params)},f(\theta)
                \text{ (objective)},                                                             \\
            &\hspace{13mm}      \eta_{+/-} \text{ (etaplus, etaminus)}, \Gamma_{max/min}
                \text{ (step sizes)}                                                             \\
            &\textbf{initialize} :   g^0_{prev} \leftarrow 0,
                \: \eta_0 \leftarrow \text{lr (learning rate)}                                   \\
            &\rule{110mm}{0.4pt}                                                                 \\
            &\textbf{for} \: t=1 \: \textbf{to} \: \ldots \: \textbf{do}                         \\
            &\hspace{5mm}g_t           \leftarrow   \nabla_{\theta} f_t (\theta_{t-1})           \\
            &\hspace{5mm} \textbf{for} \text{  } i = 0, 1, \ldots, d-1 \: \mathbf{do}            \\
            &\hspace{10mm}  \textbf{if} \:   g^i_{prev} g^i_t  > 0                               \\
            &\hspace{15mm}  \eta^i_t \leftarrow \mathrm{min}(\eta^i_{t-1} \eta_{+},
                \Gamma_{max})                                                                    \\
            &\hspace{10mm}  \textbf{else if}  \:  g^i_{prev} g^i_t < 0                           \\
            &\hspace{15mm}  \eta^i_t \leftarrow \mathrm{max}(\eta^i_{t-1} \eta_{-},
                \Gamma_{min})                                                                    \\
            &\hspace{15mm}  g^i_t \leftarrow 0                                                   \\
            &\hspace{10mm}  \textbf{else}  \:                                                    \\
            &\hspace{15mm}  \eta^i_t \leftarrow \eta^i_{t-1}                                     \\
            &\hspace{5mm}\theta_t \leftarrow \theta_{t-1}- \eta_t \mathrm{sign}(g_t)             \\
            &\hspace{5mm}g_{prev} \leftarrow  g_t                                                \\
            &\rule{110mm}{0.4pt}                                                          \\[-1.ex]
            &\bf{return} \:  \theta_t                                                     \\[-1.ex]
            &\rule{110mm}{0.4pt}                                                          \\[-1.ex]
       \end{aligned}

    For further details regarding the algorithm we refer to the paper
    `A Direct Adaptive Method for Faster Backpropagation Learning: The RPROP Algorithm
    <http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.21.1417>`_.
    aò  
    Args:
        params (iterable): iterable of parameters to optimize or dicts defining
            parameter groups
        lr (float, optional): learning rate (default: 1e-2)
        etas (Tuple[float, float], optional): pair of (etaminus, etaplus), that
            are multiplicative increase and decrease factors
            (default: (0.5, 1.2))
        step_sizes (Tuple[float, float], optional): a pair of minimal and
            maximal allowed step sizes (default: (1e-6, 50))
        z	
        z

    )r   rF   rG   r    rH   rK   rL   rM   rN   r   r   r   rI   c                C   sÔ  t | ƒD �]Ä\}}|| }|	s"|n| }|| }|| }|| }tj ¡ s€|
r€tƒ }|jj|jjkrp|jj|ks€td|› d�ƒ‚|d7 }t |¡rºt 	|¡}t 	|¡}t 	|¡}t 	|¡}|rÒ| 
| ¡ ¡ ¡ }n| 
|¡ ¡ }|
�r6| t | d¡||¡¡ | t | d¡||¡¡ | t | d¡d|¡¡ n*||| d¡< ||| d¡< d|| d¡< | |¡ ||¡ |jtjd�}|
�r¢| t | |¡d|¡¡ nd|| |¡< |j| ¡ |dd� | |¡ qd S )NúIIf capturable=True, params and state_steps must be on supported devices: Ú.r   r   r<   éÿÿÿÿ©Úvalue)Ú	enumerater7   Ú_utilsÚis_compilingr   r/   ÚtypeÚAssertionErrorr@   Zview_as_realÚmulÚcloneÚsignÚcopy_ÚwhereÚgtÚltÚeqZmul_Úclamp_rD   Zaddcmul_)r   rF   rG   r    rH   rK   rL   rM   rN   r   r   r   rI   ÚiÚparamr?   r=   r>   r,   Úcapturable_supported_devicesra   r*   r*   r+   Ú_single_tensor_rpropØ   sJ    ÿ
þ
ý




rk   c             
      s:  t | ƒdkrd S |rtdƒ‚tj ¡ s\|
r\tƒ ‰ t‡ fdd„t| |ƒD ƒƒs\tdˆ › d�ƒ‚t 	| ||||g¡}| 
¡ D �]º\\}}}}}}|d jr²tj|tjddd	�dd
� nt |d¡ |rÐt||||ƒ t ||¡}|	rêt |¡ t ||¡ |	�rt |¡ |}t |¡ |
�rv|D ]T}| t | d¡||¡¡ | t | d¡||¡¡ | t | d¡d|¡¡ �qn6|D ]0}||| d¡< ||| d¡< d|| d¡< �qzt ||¡ |D ]}| ||¡ �q¼t|ƒ}tt |ƒƒD ],}||  t ||  |¡d|| ¡¡ �qä~dd„ |D ƒ}tj|||dd� qxd S )Nr   z#_foreach ops don't support autogradc                 3   s.   | ]&\}}|j j|j jko$|j jˆ kV  qd S ©N)r/   r]   )Ú.0r;   r,   ©rj   r*   r+   Ú	<genexpr>4  s   þz&_multi_tensor_rprop.<locals>.<genexpr>rU   rV   r!   Úcpu)r/   )Úalphar   c                 S   s   g | ]}|  ¡ ‘qS r*   )ra   )rm   r?   r*   r*   r+   Ú
<listcomp>~  s     z'_multi_tensor_rprop.<locals>.<listcomp>rW   rX   )r6   r^   r7   r[   r\   r   ÚallÚzipr   Z"_group_tensors_by_device_and_dtypeÚvaluesZis_cpuZ_foreach_add_r9   r   Z_foreach_mulZ_foreach_neg_Z_foreach_copy_Z_foreach_sign_rb   rc   rd   re   rf   Z_foreach_mul_rg   ÚlistÚrangeZ_foreach_addcmul_)r   rF   rG   r    rH   rK   rL   rM   rN   r   r   r   rI   Zgrouped_tensorsZgrouped_paramsZgrouped_gradsZgrouped_prevsZgrouped_step_sizesZgrouped_state_stepsÚ_Zsignsra   r>   rh   Z
grad_signsr*   rn   r+   Ú_multi_tensor_rprop  sˆ    ý

üÿ	ú

  ÿ   ÿ


 ÿ   ÿry   )Zsingle_tensor_fnF)r   rF   rG   r    rH   r   r   r   r   rI   rK   rL   rM   rN   c
                C   s’   t j ¡ s$tdd„ |D ƒƒs$tdƒ‚|dkr>t| |dd�\}}|rTt j ¡ rTtdƒ‚|rht j ¡ sht}nt	}|| |||||
|||||||	d� dS )	zpFunctional API that performs rprop algorithm computation.

    See :class:`~torch.optim.Rprop` for details.
    c                 s   s   | ]}t |tjƒV  qd S rl   )Ú
isinstancer7   r   )rm   Útr*   r*   r+   ro   ¢  s    zrprop.<locals>.<genexpr>zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF)Z	use_fusedz6torch.jit.script not supported with foreach optimizers)rK   rL   rM   rN   r   r   r   rI   )
r7   r[   r\   rs   rB   r   ZjitZis_scriptingry   rk   )r   rF   rG   r    rH   r   r   r   r   rI   rK   rL   rM   rN   rx   Úfuncr*   r*   r+   r   ˆ  s@    ÿÿ  ÿ
ó)NFFFF)Útypingr   r   r   r7   r   Z	optimizerr   r   r	   r
   r   r   r   r   r   r   r   r   Ú__all__r   Ú__doc__r8   rS   rk   ry   r   r*   r*   r*   r+   Ú<module>   sˆ   8 #
öõôóÝÿ8òEòl	     ôï