o h5@slddlZddlmZddlmZmZmZmZmZmZm Z ddl m Z m Z ddgZ GdddeZd d ed e d ed e_   d!de ede ede ede edede ededededededefddZddZde ede ede ede edededededededefddZde ede ede ede edededededededefdd ZdS)"N)Tensor) Optimizer_use_grad_for_differentiable _get_value_default_to_fused_or_foreach_differentiable_doc _foreach_doc _maximize_doc)ListOptionalAdagradadagradcsneZdZ      dddddeeded effd d Zfd d ZddZddZe dddZ Z S)r {Gz?r绽|=NF)maximizedifferentiableforeachrrc sd|ks td|d|kstd|d|ks!td|d|ks,td|d|ks7td|t|||||||| d} t|| |jD],} | dD]%} |j| } td| d <t| rkt ||n|}tj | |tj d | d <qSqMdS) NgzInvalid learning rate: zInvalid lr_decay value: zInvalid weight_decay value: z)Invalid initial_accumulator_value value: zInvalid epsilon value: )lrlr_decayeps weight_decayinitial_accumulator_valuerrrparamsstep) memory_formatsum) ValueErrordictsuper__init__ param_groupsstatetorchtensor is_complexcomplex full_likepreserve_format)selfrrrrrrrrrdefaultsgrouppr" init_value __class__I/var/www/html/ai/venv/lib/python3.10/site-packages/torch/optim/adagrad.pyr sH      zAdagrad.__init__cst||jD]}|dd|dd|ddq t|j}t|dko3t |dd}|sF|D]}t t |d|d<q8dSdS)NrrFrrr) r __setstate__r! setdefaultlistr"valueslenr# is_tensorr$float)r)r"r+ state_valuesstep_is_tensorsr.r0r1r2?s     zAdagrad.__setstate__cCs4|jD]}|dD] }|j|}|dq qdS)Nrr)r!r" share_memory_)r)r+r,r"r0r0r1 share_memoryNs   zAdagrad.share_memoryc Cshd}|dD]+}|jdur1|jjrd}||||j|j|}||d||dq|S)NFrTrr)grad is_sparseappendr") r)r+params_with_gradgrads state_sums state_stepshas_sparse_gradr,r"r0r0r1 _init_groupTs     zAdagrad._init_groupc Csd}|durt |}Wdn1swY|jD]1}g}g}g}g}||||||}t|||||d|d|d|d||d|d|dd q |S) zPerforms a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nrrrrrrr)rrrrrErrr)r# enable_gradr!rFr) r)closurelossr+rArBrCrDrEr0r0r1rbs4  z Adagrad.step)rrrrrNN) __name__ __module__ __qualname__r boolr r2r=rFrr __classcell__r0r0r.r1r s,    3a[Implements Adagrad algorithm. .. math:: \begin{aligned} &\rule{110mm}{0.4pt} \\ &\textbf{input} : \gamma \text{ (lr)}, \: \theta_0 \text{ (params)}, \: f(\theta) \text{ (objective)}, \: \lambda \text{ (weight decay)}, \\ &\hspace{12mm} \tau \text{ (initial accumulator value)}, \: \eta\text{ (lr decay)}\\ &\textbf{initialize} : state\_sum_0 \leftarrow 0 \\[-1.ex] &\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} \tilde{\gamma} \leftarrow \gamma / (1 +(t-1) \eta) \\ &\hspace{5mm} \textbf{if} \: \lambda \neq 0 \\ &\hspace{10mm} g_t \leftarrow g_t + \lambda \theta_{t-1} \\ &\hspace{5mm}state\_sum_t \leftarrow state\_sum_{t-1} + g^2_t \\ &\hspace{5mm}\theta_t \leftarrow \theta_{t-1}- \tilde{\gamma} \frac{g_t}{\sqrt{state\_sum_t}+\epsilon} \\ &\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 `Adaptive Subgradient Methods for Online Learning and Stochastic Optimization`_. a Args: params (iterable): iterable of parameters to optimize or dicts defining parameter groups lr (float, optional): learning rate (default: 1e-2) lr_decay (float, optional): learning rate decay (default: 0) weight_decay (float, optional): weight decay (L2 penalty) (default: 0) eps (float, optional): term added to the denominator to improve numerical stability (default: 1e-10) z z .. _Adaptive Subgradient Methods for Online Learning and Stochastic Optimization: http://jmlr.org/papers/v12/duchi11a.html FrrBrCrDrErrrrrrrc Cstdd|Ds td|durt||dd\} }|r%tjr%td|r/tjs/t} nt} | ||||||| | || |d dS) ztFunctional API that performs Adagrad algorithm computation. See :class:`~torch.optim.Adagrad` for details. css|] }t|tjVqdSrJ) isinstancer#r).0tr0r0r1 szadagrad..zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizersrrrrrErr)all RuntimeErrorrr#jit is_scripting_multi_tensor_adagrad_single_tensor_adagrad)rrBrCrDrErrrrrrr_funcr0r0r1rs0 cCs8|}|dks|dkrt|St|||S)Nr)sizenumelr# empty_likesparse_coo_tensor)r> grad_indicesr5r^r0r0r1 _make_sparses rcc Cslt||||D]\} } } }|d7}t|}| s| n| } |dkr.| jr'td| j| |d} |d|d|}| jrp| } | }| }| t | || d| | }| |}| jt | |||| dqt | }|rt | } t | } t | } | j| | dd| r| |}n| |}| j| || d|rt | } t | } qdS)Nrrz;weight_decay option is not compatible with sparse gradientsalphavalue)ziprr?rWaddcoalesce_indices_valuesadd_rcpow sparse_masksqrt_r#r% view_as_realaddcmul_sqrtaddcdiv_view_as_complex)rrBrCrDrrrrrErrparamr> state_sumstep_trclrrb grad_valuesstd std_valuesr%r0r0r1r[sH       r[csj| rJdt|dkrdSt||||g} | D]\\} } }}}tdd| D}|r?t| | ||||dd| d S| rFt| } dd | D} d d |D}d d | D} t|d |dkrx| rptj| | |d ntj | | |d } fdd |D}tj || | d dt |}t|||dks| rt | || }nt | |}t| ||qdS)Nz#_foreach ops don't support autogradrcss|]}|jVqdSrJ)r?)rQr>r0r0r1rSDsz(_multi_tensor_adagrad..TFrUcS$g|]}t|rt|n|qSr0r#r%rrrQxr0r0r1 Y$z)_multi_tensor_adagrad..cSr~r0rrr0r0r1rZscSr~r0rrr0r0r1r]rrrdcs&g|]} dt|dqS)r)r)rQrrrr0r1ris&rg)r6r"_group_tensors_by_device_and_dtyper5anyr[r# _foreach_neg _foreach_add_ _foreach_add_foreach_addcmul_ _foreach_sqrt _foreach_mul_ _foreach_mul_foreach_addcdiv_)rrBrCrDrrrrrErrgrouped_tensorlists device_params device_gradsdevice_state_sumsdevice_state_stepsr\device_has_sparse_grad minus_clrr| numeratorr0rr1rZ,sT          rZ)NNF)r#r optimizerrrrrrr r typingr r __all__r __doc__rNr8rrcr[rZr0r0r0r1s $   3     5    :