U MhPO@sddlmZmZddlZddlmZddlmZddlmZm Z m Z m Z m Z m Z mZmZmZmZddgZGd ddeZd d e d e d e d e_deeeeeeeeeeeeeeeeeeeeeeeedddZddZeeeeeeeeeeeeeeeeeeeedddZeeeeeeeeeeeeeeeeeeeedddZeeeeeeeeeeeeeeeeeeeeddddZdS))ListOptionalN)Tensor)$_get_fused_kernels_supported_devices) _default_to_fused_or_foreach_differentiable_doc _foreach_doc_get_scalar_dtype _get_value _maximize_doc_use_grad_for_differentiable _view_as_real OptimizerParamsTAdagradadagradc sreZdZdddddeeeeeeeeeeeed fdd Zfd d Zd d Z ddZ e dddZ Z S)r{Gz?r绽|=NF)maximizedifferentiablefused) paramslrlr_decay weight_decayinitial_accumulator_valueepsforeachrrrc sd|kstd|d|ks,td|d|ksBtd|d|ksXtd|d|ksntd|t|||||||| | d } t|| | r| rtdd |_td tfd d |j Dstd d|rtd|j D]} | dD]z} |j | }| dr6t j dt | dd| jdnt jdt d|d<t | r`t||n|}t j| |t jd|d<qqdS)NgzInvalid learning rate: zInvalid lr_decay value: zInvalid weight_decay value: z)Invalid initial_accumulator_value value: zInvalid epsilon value: ) rrrrrrrrrz)`fused` does not support `differentiable`Tcudac3s2|]*}|dD]}|jjko&t|VqqdS)rN)devicetypetorchZis_floating_point).0ZpgpZfused_supported_devicesE/var/www/html/venv/lib/python3.8/site-packages/torch/optim/adagrad.py Gs z#Adagrad.__init__..zX`fused=True` requires all the params to be floating point Tensors of supported devices: .z0`fused` and `foreach` cannot be `True` together.rrr&Zis_fused)dtyper r+step)Z memory_formatsum) ValueErrordictsuper__init__ RuntimeErrorZ_step_supports_amp_scalingrremoveall param_groupsstater"Zzerosr r tensor is_complexcomplexZ full_likeZpreserve_format)selfrrrrrrrrrrdefaultsgroupr$r7Z init_value __class__r%r'r2st         zAdagrad.__init__cst|d}|jD]4}|dd|dd|dd|dd}qt|j}t|dkovt |dd}|s|D]$}tj t |dt |dd |d<qdS) NrrFrrrr-r*r,) r1 __setstate__r6 setdefaultlistr7valueslenr"Z is_tensorr8floatr )r;r7rr=Z state_valuesZstep_is_tensorsr>r&r'r@hs"       zAdagrad.__setstate__cCs4|jD](}|dD]}|j|}|dqqdS)Nrr.)r6r7Z share_memory_)r;r=r$r7r&r&r' share_memory}s   zAdagrad.share_memoryc Cs~d\}}|dD]d}|jdk r||jjO}|t|O}||||j|j|} || d|| dq||fS)N)FFrr.r-)grad is_sparser"r9appendr7) r;r=params_with_gradgrads state_sums state_stepshas_sparse_grad has_complexr$r7r&r&r' _init_groups      zAdagrad._init_groupc Csd}|dk r&t |}W5QRX|jD]}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) rrrrrOrrrrPrrRrS)r"Z enable_gradr6rQrgetattr) r;closureZlossr=rKrLrMrNrOrPr&r&r'r-sF    z Adagrad.step)rrrrrN)N)__name__ __module__ __qualname__rrErboolr2r@rGrQr r- __classcell__r&r&r>r'rs6 P 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`_. aH 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) 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 F)rrLrMrNrrRrSrOrrrPrrrrrc Cstdd|Dstd|dkr<|dkr.zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF)Z use_fusedz6torch.jit.script not supported with foreach optimizersz4torch.jit.script not supported with fused optimizers rrrrrOrrrPrRrS) r5r3rr"ZjitZ is_scripting_fused_adagrad_multi_tensor_adagrad_single_tensor_adagrad)rrLrMrNrrRrSrOrrrPrrrrr_funcr&r&r'rsN cCs|}t|||Sr[)sizer"Zsparse_coo_tensor)rH grad_indicesrCrdr&r&r' _make_sparse=srf)rrLrMrNrRrSrrrrrOrrrPc Cs|dkr|dkstt||||D]\\}}}}|d7}t|}| sH|n| }|dkrr|jrdtd|j||d}|d|d|}|jr|}|}|}| t ||| d| |}| | }|j t ||||| dq"t|}|r$t|}t|}t|}|j||dd| rH|| }n| | }|j||| d|r"t|}t|}q"dS)Nrrz;weight_decay option is not compatible with sparse gradientsalphavalue)AssertionErrorzipr rIr3addZcoalesceZ_indicesZ_valuesZadd_rfpowZ sparse_maskZsqrt_r"r9Z view_as_realZaddcmul_sqrtZaddcdiv_Zview_as_complex)rrLrMrNrRrSrrrrrOrrrPparamrHZ state_sumZstep_tr-ZclrreZ grad_valuesstdZ std_valuesr9r&r&r'raBsH      racs| r td|dkr|dks tt|dkr0dSt||||g}|D]J\\}}}}}| optdd|D}|rt|||||| d| | | ||dqJ| rt|||| rt |}|dj rtj |tj ddd dd n t |d |dkr| rtj |||d ntj |||d }fd d |D}tj|||d dt|}t || |dksj| r|t|||}n t||}t|||qJdS)Nz#_foreach ops don't support autogradrcss|] }|jVqdSr[)rI)r#rHr&r&r'r(sz(_multi_tensor_adagrad..Tr^g?cpu)r rgrcs&g|]} dt|dqS)r)r )r#r-rrr&r' sz)_multi_tensor_adagrad..rj)rlrDr"_group_tensors_by_device_and_dtyperCanyrarr"Z _foreach_negZis_cpu _foreach_add_r8Z _foreach_addZ_foreach_addcmul_Z _foreach_sqrtZ _foreach_mul_Z _foreach_mulZ_foreach_addcdiv_)rrLrMrNrRrSrrrrrOrrrPZgrouped_tensorlists device_params device_gradsdevice_state_sumsdevice_state_stepsrbZdevice_has_sparse_gradZ minus_clrrr numeratorr&rtr'r`s                r`)rrLrMrNrRrSrrrrrOrrrPreturncCs>|sdS| s| rtd| r$td|dk r6|j|ind}|dk rL|j|ind}t||||g}|D]\\}}\\}}}}}d\}}|dk r|dk r||kr|j|dd||<||}|dk r|dk r||kr|j|dd||<||}t|dtj|||||||| | ||d |dk rjt ||gt |qjdS)Nz5`fused` does not support sparse grad or complex paramzs  0 *   9 J > g