˲j S!ddlmZmZmZmZddlZddlmZddlmZm Z m Z m Z m Z m Z mZmZmZmZmZmZddgZGddeZd d ed e d ed e d ze_ d%deedeedeedeedeedeedeededeedededededededef dZd Zdeedeedeedeedeedeededededededededefd!Zdeedeedeedeedeedeededededededededefd"Zdeedeedeedeedeedeededededededededed#dfd$ZdS)&)castListOptionalUnionN)Tensor) _default_to_fused_or_foreach_device_dtype_check_for_fused_differentiable_doc _foreach_doc_get_scalar_dtype _get_value _maximize_doc _params_doc_use_grad_for_differentiable _view_as_real OptimizerParamsTAdagradadagradceZdZ ddddddedeeefd ed ed ed ed eedededeeffdZ fdZ dZ dZ e ddZxZS)r{Gz?r绽|=NF)maximizedifferentiablefusedparamslrlr_decay weight_decayinitial_accumulator_valueepsforeachrrrc t|tr'|dkrtdd|kstd|d|kstd|d|kstd|d|kstd|d|kstd|t |||||||| | } t || | r)| rtd |rtd d |_|j D]} | d D]} |j | }| dr0tj dt| d| jn!tjdt|d<tj| rt#||n|}tj| |tj|d<dS)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"r r!r#rrrz)`fused` does not support `differentiable`z0`fused` and `foreach` cannot be `True` together.Trris_fused)dtypedevicer(step) memory_formatsum) isinstancernumel ValueErrordictsuper__init__ RuntimeError"_need_device_dtype_check_for_fused param_groupsstatetorchzerosr r)tensor is_complexcomplex full_likepreserve_format)selfrrrr r!r"r#rrrdefaultsgrouppr7 init_value __class__s DC:\Projects\Historical\venv\Lib\site-packages\torch/optim/adagrad.pyr3zAdagrad.__init__si b& ! ! }tj t|dt| |d<=dSdS) Nr#rFrrrr+r&r*) r2 __setstate__r6 setdefaultlistr7valueslenr8 is_tensorr:floatr )r?r7rrA state_valuesstep_is_tensorsrDs rErHzAdagrad.__setstate__as@ U###& 4 4E   Y - - -   Z / / /   -u 5 5 5$$Wd33EEDJ--//00 l++q0 eo OF #7 7  !  !L!F)$$,=u,M,M,M&     rFc~|jD]4}|dD])}|j|}|d*5dS)Nrr-)r6r7 share_memory_)r?rArBr7s rE share_memoryzAdagrad.share_memoryvsZ& - -E8_ - - 1 e **,,,, - - -rFcd\}}|dD]}|j|dr)t|ddrt|dd|_||jjz}|t j|z}||||j|j|} || d|| d ||fS) N)FFrrr5T)cuda_unsupportedFr-r+) gradgetattrr r5 is_sparser8r;appendr7) r?rAparams_with_gradgrads state_sums state_stepshas_sparse_grad has_complexrBr7s rE _init_groupzAdagrad._init_group|s'3$x 2 2Av!>Dg8''D 2!dKKKK>CD;16#33u/222  ''*** QV$$$ 1 !!%,///""5=111 ++rFcd}|5tj5|}dddn #1swxYwY|jD]}g}g}g}g}||||||\}} t |||||d|d|d|d||d|d|d| |d t |d dt |d d |S) zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nrr rr"r#rrr grad_scale found_inf) rr rr"r_r#rrr`rrcrd)r8 enable_gradr6rarrX) r?closurelossrAr[r\r]r^r_r`s rEr+z Adagrad.steps`  "$$ ! !wyy ! ! ! ! ! ! ! ! ! ! ! ! ! ! !&  E-/ "$E')J(*K+/+;+;' K,, (O[  ;">2z*%L /i(z*$%56'Gn"4t<<!$ T::!     & s /33)rrrrrNN)__name__ __module__ __qualname__rrrNrrboolr3rHrTrarr+ __classcell__)rDs@rErrsJ$(+,"&D$ $DDDD %- D D  D $) DD$DDD~DDDDDDL*--- ,,,*"***"!*****rFa[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 Frr\r]r^rrcrdr_r#rr`rr rr"rc 4td|Dstd||t|| d\}}|d}|d}|r-tjrtd|r-tjrtd|r&tjst }n/|r&tjst}nt}|||||| | | |||| | ||dS) ztFunctional API that performs Adagrad algorithm computation. See :class:`~torch.optim.Adagrad` for details. c3JK|]}t|tjVdSrh)r.r8r).0ts rE zadagrad.. s.@@qz!U\**@@@@@@rFzPAPI 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 rr rr"r_rrr`rcrd) allr4r r8jit is_scripting_fused_adagrad_multi_tensor_adagrad_single_tensor_adagrad)rr\r]r^rrcrdr_r#rr`rr rr"r_funcs rErrsn2 @@K@@@ @ @  ^    }1 Ne   7 }U59))++USTTT S''))SQRRR &UY++--& &//11&$%D  ! '%rFcV|}tj|||Srh)sizer8sparse_coo_tensor)rW grad_indicesrKr~s rE _make_sparser<s$ 99;;D  "< > >>rFc ||Jt||||D]F\}}}}|dz }t|}| s|n| }|dkr-|jrtd|||}|d|dz |zzz }|jr|}|}|}|t||| d| |}| | }|t||||z | Xtj|}|r ===??D==??L,,..K NN<lKOOA.s59 9 #DN9 9 9 9 9 9 rFTrtg?cpu)r)rrcHg|]} dt|dz zzz S)r)r)rpr+rrs rE z)_multi_tensor_adagrad..sD   >BRC1 4((1,88 9   rFr)rLr"_group_tensors_by_device_and_dtyperKrrranyrzrr8 _foreach_negcompiler is_compilingis_cpu _foreach_add_r: _foreach_add_foreach_addcmul_ _foreach_sqrt _foreach_mul_ _foreach_mul_foreach_addcdiv_)rr\r]r^rcrdrr rr"r_rrr`grouped_tensorlistsdevice_params_ device_grads_device_state_sums_device_state_steps_r{ device_params device_gradsdevice_state_sumsdevice_state_stepsdevice_has_sparse_grad minus_clrr numerators ` ` rEryry~s1"DDDDD   )"3"3 3 6{{a#F  K0 & & ( ( M?M?  T&\>:: DL-88  f/ABB!$v,0CDD!0" S9 9 '39 9 9 6 6  "  "!")! $!-'%#       J -7H I I I  < -l;;L ~**,, 71CA1F1M 7  "ELU$C$C$C3        2A 6 6 6 1   #L-|TTTTT$1 -|         FX     1<UVWWWW!"344 C%%% 1      i 8 8 8$II*<CCI  y#>>>>[M?M?rFreturncr|sdS| s| rtd| rtd| |j|ind}| |j|ind}tj||||g}|D]F\\}}\\}}}}}t t t|}t t t|}t t t|}t t t|}d\}}|(|&||vr||d||<||}|(|&||vr||d||<||}tj |dtj |||||||| | || |&tj ||gt|zHdS)Nz5`fused` does not support sparse grad or complex paramz:: DL-88  f/ABB!$v,0CDD.8++  !o&A_,,*4--T-*R*R' / 7   ^%?..)2f4)P)Pv&-f5  .222     %(&   '  "%5$6=O9P9P$P   M((rF)NNNFNFF)typingrrrrr8r optimizerr r r r r rrrrrrr__all__r__doc__rlrNrrrzryrxr%rFrErs............                              i bbbbbibbbL4    5.p!#'"&"" GG LG <GV Gf G D> G  GGGd^GGG G !G"#G$ %G&'GGGGT??? := L:= <:=V :=f :=  :=  := :=:=:= :=:=:=:=:=:=:=:=zh? Lh? <h?V h?f h?  h?  h? h?h?h? h?h?h?h?h?h?h?h?VK LK <KV Kf K  K  K KKK KKKKK  !KKKKKKrF