o inR#@s<ddlmZddlZddlmZddlmZmZmZmZm Z m Z m Z m Z m Z mZmZmZmZddgZGdddeZd d e d ed e d ed e_       d)deedeedeedeededBdedBdedBdededBdedededededededdf"d dZd!d"ZdeedeedeedeededBdedBdededededededededdfd#d$ZdeedeedeedeededBdedBdededededededededdfd%d&ZdeedeedeedeededBdedBdededededededededdfd'd(ZdS)*)castN)Tensor) _default_to_fused_or_foreach_device_dtype_check_for_fused_differentiable_doc _foreach_doc_get_scalar_dtype _get_value _maximize_doc _params_doc _to_scalar_use_grad_for_differentiable _view_as_real OptimizerParamsTAdagradadagradcseZdZ      ddddddedeeBd ed ed ed ed edBdedededBddffddZfddZdddZ ddZ e dddZ Z S)r{Gz?r绽|=NF)maximizedifferentiablefusedparamslrlr_decay weight_decayinitial_accumulator_valueepsforeachrrrreturnc sbt|tr|dkrtdd|kstd|d|ks%td|d|ks0td|d|ks;td|d|ksFtd||||||||| | d } t|| | rj| ratd |rgtd d |_|jD]A} | d D]:} |j | }| drt j dt | dd| j dnt jdt d|d<t | rt||n|}t j| |t jd|d<qsqmdS)NrzTensor lr must be 1-elementgzInvalid learning rate: zInvalid lr_decay value: zInvalid weight_decay value: z)Invalid initial_accumulator_value value: zInvalid epsilon value: ) rrrrrrrrrz)`fused` does not support `differentiable`z0`fused` and `foreach` cannot be `True` together.Trris_fused)dtypedevicer$step) memory_formatsum) isinstancernumel ValueErrorsuper__init__ RuntimeError"_need_device_dtype_check_for_fused param_groupsstatetorchzerosr r%tensor is_complexcomplex full_likepreserve_format)selfrrrrrrrrrrdefaultsgrouppr2 init_value __class__r!U/sda-disk/www/egybert/egybert_env/lib/python3.10/site-packages/torch/optim/adagrad.pyr.sf       zAdagrad.__init__cst|d}|jD]}|dd|dd|dd|dd}q t|j}t|dko;t |dd}|sS|D]}tj t |dt |dd |d<q@dSdS) NrrFrrrr'r"r&) r- __setstate__r1 setdefaultlistr2valueslenr3 is_tensorr5floatr )r:r2rr< state_valuesstep_is_tensorsr?r!rArBbs$       zAdagrad.__setstate__cCs4|jD]}|dD] }|j|}|dq qdS)z6Calls tensor.share_memory_() on the state sum tensors.rr)N)r1r2 share_memory_)r:r<r=r2r!r!rA share_memoryws   zAdagrad.share_memoryc Csd\}}|dD]E}|jdurM|dr"t|ddr"t|ddd|_||jjO}|t|O}||||j|j|} || d|| d q||fS) N)FFrrr0T)cuda_unsupportedFr)r') gradgetattrrr0 is_sparser3r6appendr2) r:r<params_with_gradgrads state_sums state_stepshas_sparse_grad has_complexr=r2r!r!rA _init_group~s&        zAdagrad._init_groupc Csd}|durt |}Wdn1swY|jD]A}g}g}g}g}||||||\}} t|||||d|d|d|d||d|d|d| |d t|d dt|d dd q |S) zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nrrrrrrrr grad_scale found_inf) rrrrrWrrrrXrrZr[)r3 enable_gradr1rYrrP) r:closurelossr<rSrTrUrVrWrXr!r!rAr's@     z Adagrad.step)rrrrrN)r NN)__name__ __module__ __qualname__rrHrboolr.rBrMrYrr' __classcell__r!r!r?rArsN      F a[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 \tau \\[-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`_. z Args: a lr (float, Tensor, 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) initial_accumulator_value (float, optional): initial value of the sum of squares of gradients (default: 0) eps (float, optional): term added to the denominator to improve numerical stability (default: 1e-10) z a fused (bool, optional): whether the fused implementation (CPU only) is used. Currently, `torch.float64`, `torch.float32`, `torch.float16`, and `torch.bfloat16` are supported. (default: None). Please note that the fused implementations does not support sparse or complex gradients. .. _Adaptive Subgradient Methods for Online Learning and Stochastic Optimization: http://jmlr.org/papers/v12/duchi11a.html FrrTrUrVrrZr[rWrrrXrrrrrr c Cstdd|Ds td|dur|durt|| dd\}}|dur$d}|dur*d}|r5tjr5td|r@tjr@td|rJtjsJt}n |rTtjsTt}nt}|||||| | | |||| | ||d dS) ztFunctional API that performs Adagrad algorithm computation. See :class:`~torch.optim.Adagrad` for details. css|] }t|tjVqdSr_)r*r3r).0tr!r!rA szadagrad..zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizersz4torch.jit.script not supported with fused optimizers rrrrrWrrrXrZr[) allr/rr3jit is_scripting_fused_adagrad_multi_tensor_adagrad_single_tensor_adagrad)rrTrUrVrrZr[rWrrrXrrrrr_funcr!r!rArsJ  cCs|}t|||Sr_)sizer3sparse_coo_tensor)rO grad_indicesrErrr!r!rA _make_sparse>sruc Cs|dus|dur tdtjst|}t||||ddD]\}}}}|d7}t|}| s0|n| }|dkrE|jr>td|j ||d}|d|d|}|jr| }| }| }| t|||d||}|  | }|j t||||| dqt|}|rt|}t|}t|}|j||dd | r|| }n| | }|j||| d |rt|}t|}qdS) N,Expected grad_scale and found_inf to be NoneT)strictrrz;weight_decay option is not compatible with sparse gradientsalphavalue)AssertionErrorr3rkrlr zipr rQr/addcoalesce_indices_valuesadd_rupow sparse_masksqrt_r6 view_as_realaddcmul_sqrtaddcdiv_view_as_complex)rrTrUrVrZr[rrrrrWrrrXparamrO state_sumstep_tr'clrrt grad_valuesstd std_valuesr6r!r!rAroCsT         rocs| rtd|dus|durtdt|dkrdStt||||g}|D]\\}}}}}ttt|}ttt|}ttt|}ttt|}| oYt dd|D}|rot |||||| d| | | ||dq+| rwt |||| r~t |}t js|djrt j|t jdd d dd nt |d |dkr| rt j|||d nt j|||d }fd d|D}t j|||d dt |}t || |dks| rt |||}nt ||}t |||q+dS)Nz#_foreach ops don't support autogradrvrcss|]}|jVqdSr_)rQ)rerOr!r!rArgs z(_multi_tensor_adagrad..Trig?cpu)r%rxrcs&g|]} dt|dqS)r)r )rer'rrr!rA sz)_multi_tensor_adagrad..r{)r}rFr r"_group_tensors_by_device_and_dtyperErrDranyrorr3 _foreach_negcompiler is_compilingis_cpu _foreach_add_r5 _foreach_add_foreach_addcmul_ _foreach_sqrt _foreach_mul_ _foreach_mul_foreach_addcdiv_)rrTrUrVrZr[rrrrrWrrrXgrouped_tensorlistsdevice_params_ device_grads_device_state_sums_device_state_steps_rp device_params device_gradsdevice_state_sumsdevice_state_stepsdevice_has_sparse_grad minus_clrr numeratorr!rrArns            rncCs~|sdS| s| r td| rtdt|}|dur|j|ind}|dur*|j|ind}t||||g}|D]\\}}\\}}}}}ttt|}ttt|}ttt|}ttt|}d\}}|dur~|dur~||vrz|j |dd||<||}|dur|dur||vr|j |dd||<||}t |dt j |||||||| | ||d |durt ||gt|q9dS)Nz5`fused` does not support sparse grad or complex paramzs6  <(   8     J      D      o