a hTQ@sddlmZmZmZddlZddlmZddlmZmZm Z m Z m Z m Z m Z mZmZmZmZmZmZddgZGdddeZd d ed e d e d e d e_deeeeeeeeeeeeeeeeeeeeeeeedddZddZeeeeeeeeeeeeeeeeeeeedddZeeeeeeeeeeeeeeeeeeeedddZeeeeeeeeeeeeeeeeeeeeddddZdS))castOptionalUnionN)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 OptimizerParamsTAdagradadagradc szeZdZdddddeeeefeeeeeeeeeed fdd Z fd d Z d d Z ddZ e dddZZS)r{Gz?r绽|=NF)maximizedifferentiablefused) paramslrlr_decay weight_decayinitial_accumulator_valueepsforeachrrrc sht|tr|dkrtdd|ks4td|d|ksJtd|d|ks`td|d|ksvtd|d|kstd|t|||||||| | d } t|| | r| rtd |rtd d |_|j D]} | d D]x} |j | }| drt j dt | dd| jdnt jdt d|d<t | rFt||n|}t j| |t jd|d<qqdS)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: ) rrr rrr!rrrz)`fused` does not support `differentiable`z0`fused` and `foreach` cannot be `True` together.TrrZis_fused)dtypedevicer$step)Z memory_formatsum) isinstancerZnumel ValueErrordictsuper__init__ RuntimeError"_need_device_dtype_check_for_fused param_groupsstatetorchzerosr r%tensor is_complexcomplexZ full_likeZpreserve_format)selfrrrrrr r!rrrdefaultsgrouppr1Z init_value __class__r"A/var/www/auris/lib/python3.9/site-packages/torch/optim/adagrad.pyr-sd      zAdagrad.__init__cst|d}|jD]4}|dd|dd|dd|dd}qt|j}t|dkovt |dd}|s|D]$}tj t |dt |dd |d<qdS) Nr!rFrrrr'r#r&) r, __setstate__r0 setdefaultlistr1valueslenr2Z is_tensorr4floatr )r7r1rr9Z state_valuesZstep_is_tensorsr;r"r=r>bs       zAdagrad.__setstate__cCs4|jD](}|dD]}|j|}|dqqdS)Nrr()r0r1Z share_memory_)r7r9r:r1r"r"r= share_memoryws   zAdagrad.share_memoryc Csd\}}|dD]}|jdur|drDt|ddrDt|ddd|_||jjO}|t|O}||||j|j|} || d|| d q||fS) N)FFrrr/T)Zcuda_unsupportedFr(r') gradgetattrrr/ is_sparser2r5appendr1) r7r9params_with_gradgrads state_sums state_stepshas_sparse_grad has_complexr:r1r"r"r= _init_group}s$        zAdagrad._init_groupc Csd}|dur:t|}Wdn1s00Y|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. Nrrrr r!rrr grad_scale found_inf) rrrr rNr!rrrOrrQrR)r2Z enable_gradr0rPrrG) r7closureZlossr9rJrKrLrMrNrOr"r"r=r's> $    z Adagrad.step)rrrrrN)N)__name__ __module__ __qualname__rrrCrrboolr-r>rErPrr' __classcell__r"r"r;r=rs6  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 F)rrKrLrMrrQrRrNr!rrOrrrr rc Cstdd|Dstd|dur<|dur zadagrad..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 rrrr rNrrrOrQrR) allr.rr2jit is_scripting_fused_adagrad_multi_tensor_adagrad_single_tensor_adagrad)rrKrLrMrrQrRrNr!rrOrrrr r_funcr"r"r=rsJ cCs|}t|||SrY)sizer2Zsparse_coo_tensor)rF grad_indicesrArgr"r"r= _make_sparse=sri)rrKrLrMrQrRrrrr rNrrrOc Cs|dur|dusJtjs&t|}t||||D]^\}}}}|d7}t|}| sZ|n| }|dkr|jrvtd|j||d}|d|d|}|jr | }| }| }| t |||d||}|  | }|j t ||||| dq4t|}|r8t|}t|}t|}|j||dd| r\|| }n| | }|j||| d|r4t|}t|}q4dS)Nrrz;weight_decay option is not compatible with sparse gradientsalphavalue)r2r`rarzipr rHr.addZcoalesceZ_indicesZ_valuesZadd_ripowZ sparse_maskZsqrt_r5Z view_as_realZaddcmul_sqrtZaddcdiv_Zview_as_complex)rrKrLrMrQrRrrrr rNrrrOparamrFZ state_sumZstep_tr'ZclrrhZ grad_valuesstdZ std_valuesr5r"r"r=rdBsJ       rdcs| r Jd|dur|dus Jt|dkr0dStt||||g}|D]\\}}}}}ttt|}ttt|}ttt|}ttt|}| otdd|D}|rt |||||| d| | | ||dqR| rt |||| rt |}t j s.|djr.t j|t jddd dd n t |d |dkrl| r\t j|||d nt j|||d }fd d |D}t j|||d dt |}t || |dks| rt |||}n t ||}t |||qRdS)Nz#_foreach ops don't support autogradrcss|] }|jVqdSrY)rH)rZrFr"r"r=r\sz(_multi_tensor_adagrad..Tr^g?cpu)r%rjrcs&g|]} dt|dqS)r)r )rZr'rrr"r= sz)_multi_tensor_adagrad..rm)rBrr"_group_tensors_by_device_and_dtyperArr@ranyrdrr2Z _foreach_negcompilerZ is_compilingZis_cpu _foreach_add_r4Z _foreach_addZ_foreach_addcmul_Z _foreach_sqrtZ _foreach_mul_Z _foreach_mulZ_foreach_addcdiv_)rrKrLrMrQrRrrrr rNrrrOZgrouped_tensorlistsdevice_params_ device_grads_device_state_sums_device_state_steps_re device_params device_gradsdevice_state_sumsdevice_state_stepsZdevice_has_sparse_gradZ minus_clrrt numeratorr"rvr=rcs              rc)rrKrLrMrQrRrrrr rNrrrOreturncCs|sdS| s| rtd| r$tdt|}|dur>|j|ind}|durT|j|ind}t||||g}|D] \\}}\\}}}}}ttt|}ttt|}ttt|}ttt|}d\}}|dur|dur||vr|j |dd||<||}|dur6|dur6||vr.|j |dd||<||}t |dt j |||||||| | ||d |durrt ||gt|qrdS)Nz5`fused` does not support sparse grad or complex paramzs <'   8 J B n