o
    ,izE                     @   sT   d dl mZ d dl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 )    )OptionalN   )	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   O/home/ubuntu/SoloSpeech/.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rJ   c                       s   e Zd ZdZ							d ded	e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): learning rate (default: 1)
        max_iter (int): maximal number of iterations per optimization step
            (default: 20)
        max_eval (int): maximal number of function evaluations per optimization
            step (default: max_iter * 1.25).
        tolerance_grad (float): termination tolerance on first order optimality
            (default: 1e-7).
        tolerance_change (float): termination tolerance on function
            value/parameter changes (default: 1e-9).
        history_size (int): update history size (default: 100).
        line_search_fn (str): either 'strong_wolfe' or None (default: None).
    r      NHz>r   d   paramslrmax_itermax_evaltolerance_gradr2   history_sizeline_search_fnc	           
   	      sh   |d u r
|d d }t |||||||d}	t ||	 t| jdkr'td| jd d | _d | _d S )N      )rO   rP   rQ   rR   r2   rS   rT   r   z>LBFGS doesn't support per-parameter options (parameter groups)r   rN   )dictsuper__init__lenparam_groups
ValueError_params_numel_cache)
selfrN   rO   rP   rQ   rR   r2   rS   rT   defaults	__class__r   r   rY      s$   	
zLBFGS.__init__c                 C   s&   | j d u rtdd | jD | _ | j S )Nc                 s   s.    | ]}t |rd |  n| V  qdS )r   N)r&   
is_complexnumel.0pr   r   r   	<genexpr>   s
    
zLBFGS._numel.<locals>.<genexpr>)r^   sumr]   r_   r   r   r   _numel   s
   

zLBFGS._numelc                 C   s   g }| j D ]6}|jd u r||  }n|jjr#|j d}n|jd}t	|r6t
|d}|| qt|dS )Nr#   r   )r]   gradnewrd   zero_	is_sparseto_denseviewr&   rc   view_as_realappendcat)r_   viewsrg   rq   r   r   r   _gather_flat_grad   s   


zLBFGS._gather_flat_gradc                 C   sh   d}| j D ]$}t|rt|}| }|j||||  ||d ||7 }q||  ks2J d S )Nr   alpha)r]   r&   rc   rr   rd   add_view_asrk   )r_   	step_sizeupdateoffsetrg   rd   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'   re   r   r   r   
<listcomp>  s    z&LBFGS._clone_param.<locals>.<listcomp>)r]   rj   r   r   r   _clone_param     zLBFGS._clone_paramc                 C   s$   t | j|D ]	\}}|| qd S N)zipr]   copy_)r_   params_datarg   pdatar   r   r   
_set_param  s   zLBFGS._set_paramc                 C   s0   |  || t| }|  }| | ||fS r   )r~   floatrv   r   )r_   closurer*   r+   r,   loss	flat_gradr   r   r   _directional_evaluate  s
   

zLBFGS._directional_evaluatec           &         s  t jdks	J t   jd }|d }|d }|d }|d }|d }|d }|d	 }	jjd  }
|
d
d |
dd   }t|}d}|
d
  d7  <  }|	 
 |k}|re|S |
d}|
d}|
d}|
d}|
d}|
d}|
d}|
d}d}||k r1|d7 }|
d  d7  < |
d dkr| }g }g }g }d}n||}||}||}|dkrt ||	kr|d |d |d || || |d|  ||| }t |}d|
vrdg|	 |
d< |
d }| }t|d ddD ]}|| |||  ||< |j|| ||  d qt|| }} t|D ]}|| | ||  }!| j|| || |! d q<|du re|jtjd}n|| |}|
d dkrtdd|	   | }n|}||}"|"| krnd}#|dur|dkrtd }$ 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   rO   rP   rQ   rR   r2   rT   rS   
func_evalsn_iterr,   r+   old_dirsold_stpsroH_diagprev_flat_grad	prev_lossg|=g      ?alNr#   rw   r    strong_wolfez only 'strong_wolfe' is supportedc                    s     | ||S r   )r   )r*   r+   r,   r   r_   r   r   r)     r   zLBFGS.step.<locals>.obj_func)rZ   r[   r&   enable_gradstater]   
setdefaultr   rv   r$   r   getnegsubmulr(   poprs   rangery   r%   r'   r   r
   ri   RuntimeErrorr   rJ   r~   )&r_   r   grouprO   rP   rQ   rR   r2   rT   rS   r   	orig_lossr   current_evalsr   opt_condr,   r+   r   r   r   r   r   r   r   ysysnum_oldr   qirbe_ir/   r7   x_initr)   r   r   r   step&  s   



























  z
LBFGS.step)r   rK   NrL   r   rM   N)__name__
__module____qualname____doc__r   r   intr   strrY   rk   rv   r~   r   r   r   r&   no_gradr   __classcell__r   r   ra   r   r      sD    $	 	r   )r   r   r   r   )
typingr   r&   	optimizerr   r   __all__r   rJ   r   r   r   r   r   <module>   s   

 