o
    hF                     @   sd   d dl mZmZ d dlZd dlmZ ddlmZmZ dgZdddZ		dddZ
G dd deZdS )    )OptionalUnionN)Tensor   )	OptimizerParamsTLBFGSc                 C   s   |d ur	|\}}n| |kr| |fn|| f\}}|| d||  | |   }	|	d ||  }
|
dkrj|
  }| |krN|||  || |	 || d|     }n| | | || |	 || d|     }tt|||S || d S )N      r   g       @)sqrtminmax)x1f1g1x2f2g2bounds
xmin_bound
xmax_boundd1	d2_squared2min_pos r   e/var/www/html/construction_image-detection-poc/venv/lib/python3.10/site-packages/torch/optim/lbfgs.py_cubic_interpolate   s   
	*(r   -C6??&.>   c           !   	   C   s   |   }|jtjd}| |||\}}d}||}d|||f\}}}}d}d}||
k r|||| |  ks@|dkrV||krV||g}||g}||jtjdg}||g}npt || | krk|g}|g}|g}d}n[|dkr||g}||g}||jtjdg}||g}nA|d||   }|d }|}t||||||||fd}|}|}|jtjd}|}| |||\}}|d7 }||}|d7 }||
k s.||
krd|g}||g}||g}d}|d |d	 krd
nd\}}|s||
k rt |d |d  | |	k rn t|d |d |d |d |d |d }dt|t|  } tt|| |t| | k rb|s>|t|ks>|t|kr_t |t| t |t| k rVt||  }nt||  }d}nd}nd}| |||\}}|d7 }||}|d7 }|||| |  ks||| kr|||< |||< |jtjd||< |||< |d |d krd
nd\}}nGt || | krd}n%||| ||   dkr|| ||< || ||< || ||< || ||< |||< |||< |jtjd||< |||< |s||
k s|| }|| }|| }||||fS )Nmemory_formatr   r   FTg{Gz?
   )r   )r   r   )r   r   g?)absr   clonetorchcontiguous_formatdotr   r   )!obj_funcxtdfggtdc1c2tolerance_changemax_lsd_normf_newg_newls_func_evalsgtd_newt_prevf_prevg_prevgtd_prevdonels_iterbracket	bracket_f	bracket_gbracket_gtdmin_stepmax_steptmpinsuf_progresslow_poshigh_posepsr   r   r   _strong_wolfe)   s   
$
* ""
$ DrL   c                       s   e Zd ZdZ							d ded	eeef d
ede	e dededede	e
 f fddZdd Zdd Zdd Zdd Zdd Zdd Ze dd Z  ZS )!r   a  Implements L-BFGS algorithm.

    Heavily inspired by `minFunc
    <https://www.cs.ubc.ca/~schmidtm/Software/minFunc.html>`_.

    .. warning::
        This optimizer doesn't support per-parameter options and parameter
        groups (there can be only one).

    .. warning::
        Right now all parameters have to be on a single device. This will be
        improved in the future.

    .. note::
        This is a very memory intensive optimizer (it requires additional
        ``param_bytes * (history_size + 1)`` bytes). If it doesn't fit in memory
        try reducing the history size, or use a different algorithm.

    Args:
        params (iterable): iterable of parameters to optimize. Parameters must be real.
        lr (float, optional): learning rate (default: 1)
        max_iter (int, optional): maximal number of iterations per optimization step
            (default: 20)
        max_eval (int, optional): maximal number of function evaluations per optimization
            step (default: max_iter * 1.25).
        tolerance_grad (float, optional): termination tolerance on first order optimality
            (default: 1e-7).
        tolerance_change (float, optional): termination tolerance on function
            value/parameter changes (default: 1e-9).
        history_size (int, optional): update history size (default: 100).
        line_search_fn (str, optional): either 'strong_wolfe' or None (default: None).
    r      NHz>r    d   paramslrmax_itermax_evaltolerance_gradr4   history_sizeline_search_fnc	           
   	      s   t |tr| dkrtdd|kstd| |d u r$|d d }t|||||||d}	t ||	 t| jdkrAtd| jd	 d
 | _	d | _
d S )Nr   zTensor lr must be 1-elementg        zInvalid learning rate:       )rQ   rR   rS   rT   r4   rU   rV   z>LBFGS doesn't support per-parameter options (parameter groups)r   rP   )
isinstancer   numel
ValueErrordictsuper__init__lenparam_groups_params_numel_cache)
selfrP   rQ   rR   rS   rT   r4   rU   rV   defaults	__class__r   r   r^      s,   	
zLBFGS.__init__c                 C   s&   | j d u rtdd | jD | _ | j S )Nc                 s   s.    | ]}t |rd |  n| V  qdS )r
   N)r(   
is_complexrZ   .0pr   r   r   	<genexpr>   s
    
zLBFGS._numel.<locals>.<genexpr>)rb   sumra   rc   r   r   r   _numel   s
   

zLBFGS._numelc                 C   s   g }| j D ]6}|jd u r||  }n|jjr#|j d}n|jd}t	|r6t
|d}|| qt|dS )Nr%   r   )ra   gradnewrZ   zero_	is_sparseto_denseviewr(   rg   view_as_realappendcat)rc   viewsrj   rt   r   r   r   _gather_flat_grad  s   


zLBFGS._gather_flat_gradc                 C   sh   d}| j D ]$}t|rt|}| }|j||||  ||d ||7 }q||  ks2J d S )Nr   alpha)ra   r(   rg   ru   rZ   add_view_asrn   )rc   	step_sizeupdateoffsetrj   rZ   r   r   r   	_add_grad  s   


 
zLBFGS._add_gradc                 C   s   dd | j D S )Nc                 S   s   g | ]	}|j tjd qS )r"   )r'   r(   r)   rh   r   r   r   
<listcomp>   s    z&LBFGS._clone_param.<locals>.<listcomp>)ra   rm   r   r   r   _clone_param     zLBFGS._clone_paramc                 C   s$   t | j|D ]	\}}|| qd S N)zipra   copy_)rc   params_datarj   pdatar   r   r   
_set_param"  s   zLBFGS._set_paramc                 C   s0   |  || t| }|  }| | ||fS r   )r   floatry   r   )rc   closurer,   r-   r.   loss	flat_gradr   r   r   _directional_evaluate&  s
   

zLBFGS._directional_evaluatec           &         s  t jdks	J t   jd }|d }|d }|d }|d }|d }|d }|d	 }	jjd  }
|
d
d |
dd   }t|}d}|
d
  d7  <  }|	 
 |k}|re|S |
d}|
d}|
d}|
d}|
d}|
d}|
d}|
d}d}||k r1|d7 }|
d  d7  < |
d dkr| }g }g }g }d}n||}||}||}|dkrt ||	kr|d |d |d || || |d|  ||| }t |}d|
vrdg|	 |
d< |
d }| }t|d ddD ]}|| |||  ||< |j|| ||  d qt|| }} t|D ]}|| | ||  }!| j|| || |! d q<|du re|jtjd}n|| |}|
d dkrtdd|	   | }n|}||}"|"| krnd}#|dur|dkrtd }$ fdd}%t|%|$|||||"\}}}}#|| |	 
 |k}n3|| ||krt  t  }W d   n	1 sw   Y   }|	 
 |k}d}#||#7 }|
d
  |#7  < ||krn%||krn|rn||	 
 |kr#nt	|| |k r-n||k s||
d< ||
d< ||
d< ||
d< ||
d< ||
d< ||
d< ||
d< |S )zPerform a single optimization step.

        Args:
            closure (Callable): A closure that reevaluates the model
                and returns the loss.
        r   r   rQ   rR   rS   rT   r4   rV   rU   
func_evalsn_iterr.   r-   old_dirsold_stpsroH_diagprev_flat_grad	prev_lossg|=g      ?alNr%   rz   r"   strong_wolfez only 'strong_wolfe' is supportedc                    s     | ||S r   )r   )r,   r-   r.   r   rc   r   r   r+     r   zLBFGS.step.<locals>.obj_func)r_   r`   r(   enable_gradstatera   
setdefaultr   ry   r&   r   getnegsubmulr*   poprv   ranger|   r'   r)   r   r   rl   RuntimeErrorr   rL   r   )&rc   r   grouprQ   rR   rS   rT   r4   rV   rU   r   	orig_lossr   current_evalsr   opt_condr.   r-   r   r   r   r   r   r   r   ysysnum_oldr   qirbe_ir1   r9   x_initr+   r   r   r   step-  s   



























  z
LBFGS.step)r   rM   NrN   r    rO   N)__name__
__module____qualname____doc__r   r   r   r   intr   strr^   rn   ry   r   r   r   r   r(   no_gradr   __classcell__r   r   re   r   r      sD    $
	$	r   )r   r   r    r!   )typingr   r   r(   r   	optimizerr   r   __all__r   rL   r   r   r   r   r   <module>   s   

 