ёi4SSKJr SSKrSSKJr SSKJr SSKJrJ r SSK J r SSK J r S S KJr \(a5SS KJr SSKJr SS KJr SS KJr SSKJr SSKJr S SKJr "SS\5r/r"SS\5rg)) annotationsN) TYPE_CHECKING) NotRequired)_C_opspir) framework)in_dynamic_or_pir_mode) Optimizer)Sequence)Tensor)GradientClipBase) LRScheduler)WeightDecayRegularizer)_ParameterConfigc>\rSrSr%S\S'S\S'S\S'S\S'Srg ) _RMSPropParameterConfig(zNotRequired[float]epsilonmomentumrhozNotRequired[bool]centeredN)__name__ __module__ __qualname____firstlineno____annotations____static_attributes__rX/var/www/html/banglarbhumi/venv/lib/python3.13/site-packages/paddle/optimizer/rmsprop.pyrr(s##$$ ##r!rc^\rSrSrSrSrSrSrS S U4SjjjrSr Sr S r S r U=r $) RMSProp2aw Root Mean Squared Propagation (RMSProp) is an unpublished, adaptive learning rate method. The original slides proposed RMSProp: Slide 29 of http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf . The original equation is as follows: .. math:: r(w, t) & = \rho r(w, t-1) + (1 - \rho)(\nabla Q_{i}(w))^2 w & = w - \frac{\eta} {\sqrt{r(w,t) + \epsilon}} \nabla Q_{i}(w) The first equation calculates moving average of the squared gradient for each weight. Then dividing the gradient by :math:`sqrt{v(w,t)}`. In some cases, adding a momentum term :math: `\\beta` is beneficial. In our implementation, Nesterov momentum is used: .. math:: r(w, t) & = \rho r(w, t-1) + (1 - \rho)(\nabla Q_{i}(w))^2 v(w, t) & = \beta v(w, t-1) + \frac{\eta} {\sqrt{r(w,t) + \epsilon}} \nabla Q_{i}(w) w & = w - v(w, t) if centered is True: .. math:: r(w, t) & = \rho r(w, t-1) + (1 - \rho)(\nabla Q_{i}(w))^2 g(w, t) & = \rho g(w, t-1) + (1 - \rho)\nabla Q_{i}(w) v(w, t) & = \beta v(w, t-1) + \frac{\eta} {\sqrt{r(w,t) - (g(w, t))^2 + \epsilon}} \nabla Q_{i}(w) w & = w - v(w, t) where, :math:`\rho` is a hyperparameter and typical values are 0.9, 0.95 and so on. :math:`\beta` is the momentum term. :math:`\epsilon` is a smoothing term to avoid division by zero, usually set somewhere in range from 1e-4 to 1e-8. Parameters: learning_rate (float|LRScheduler): The learning rate used to update ``Parameter``. It can be a float value or a LRScheduler. rho(float, optional): rho is :math:`\rho` in equation, default is 0.95. epsilon(float, optional): :math:`\epsilon` in equation is smoothing term to avoid division by zero, default is 1e-6. momentum(float, optional): :math:`\beta` in equation is the momentum term, default is 0.0. centered(bool, optional): If True, gradients are normalized by the estimated variance of the gradient; if False, by the uncentered second moment. Setting this to True may help with training, but is slightly more expensive in terms of computation and memory. Defaults to False. parameters (list|tuple|None, optional): List/Tuple of ``Tensor`` to update to minimize ``loss``. This parameter is required in dygraph mode. And you can specify different options for different parameter groups such as the learning rate, weight decay, etc, then the parameters are list of dict. Note that the learning_rate in parameter groups represents the scale of base learning_rate. The default value is None in static graph mode, at this time all parameters will be updated. weight_decay (int|float|WeightDecayRegularizer|None, optional): The strategy of regularization. It can be a int or float value as coeff of L2 regularization or \ :ref:`api_paddle_regularizer_L1Decay`, :ref:`api_paddle_regularizer_L2Decay`. If a parameter has set regularizer using :ref:`api_paddle_ParamAttr` already, the regularization setting here in optimizer will be ignored for this parameter. Otherwise, the regularization setting here in optimizer will take effect. Default None, meaning there is no regularization. grad_clip (GradientClipBase|None, optional): Gradient clipping strategy, it's an instance of some derived class of ``GradientClipBase`` . There are three clipping strategies ( :ref:`api_paddle_nn_ClipGradByGlobalNorm` , :ref:`api_paddle_nn_ClipGradByNorm` , :ref:`api_paddle_nn_ClipGradByValue` ). Default None, meaning there is no gradient clipping. name (str|None, optional): Normally there is no need for user to set this property. For more information, please refer to :ref:`api_guide_Name`. The default value is None. Examples: .. code-block:: python >>> import paddle >>> inp = paddle.rand([10,10], dtype="float32") >>> linear = paddle.nn.Linear(10, 10) >>> out = linear(inp) >>> loss = paddle.mean(out) >>> rmsprop = paddle.optimizer.RMSProp( ... learning_rate=0.1, ... parameters=linear.parameters(), ... weight_decay=0.01 ... ) >>> out.backward() >>> rmsprop.step() >>> rmsprop.clear_grad() >>> # Note that the learning_rate of linear_2 is 0.01. >>> linear_1 = paddle.nn.Linear(10, 10) >>> linear_2 = paddle.nn.Linear(10, 10) >>> inp = paddle.uniform(shape=[10, 10], min=-0.1, max=0.1) >>> out = linear_1(inp) >>> out = linear_2(out) >>> loss = paddle.mean(out) >>> rmsprop = paddle.optimizer.RMSProp( ... learning_rate=0.1, ... parameters=[{ # type: ignore ... 'params': linear_1.parameters() ... }, { ... 'params': linear_2.parameters(), ... 'weight_decay': 0.001, ... 'learning_rate': 0.1 ... }], ... weight_decay=0.01 ... ) >>> out.backward() >>> rmsprop.step() >>> rmsprop.clear_grad() r mean_square mean_gradc p>Uc [S5eUc [S5eUc [S5eUc [S5eSU::d [S5eSU::d [S5eSU::d [S5e[T U] UUUUU S 9 S UlX lX0lX@lXPlS Ul0Ul UUUUS .Ul g) Nzlearning_rate is not set.zrho is not set.zepsilon is not set.zmomentum is not set.z.Invalid value of epsilon, expect epsilon >= 0.z0Invalid value of momentum, expect momentum >= 0.z&Invalid value of rho, expect rho >= 0.) learning_rate parameters weight_decay grad_clipnamermspropF)rrrr) ValueErrorsuper__init__type_rho_epsilon _momentum _centered_multi_precision_master_weights _default_dict) selfr*rrrrr+r,r-r. __class__s r"r2RMSProp.__init__s  89 9 ;./ / ?23 3  34 4g~MN NhOP PczEF F '!%     !! %!   r!c [U[R[R45(d [ S5e[U[ 5(aUR S5nUGHnURUR;aM UR(aURUR5(aURU5nURURU5 URURU5 URUR U5 URR#UR5 MURUR5(a'UR(d[$R&"S5 URURU5 URURU5 URUR U5 URR#UR5 GM g)Nblock is not instance of Block.paramszAccumulating with FP16 in optimizer can lead to poor accuracy or slow convergence.Consider using multi_precision=True option of the Lars optimizer.) isinstancer Blockr TypeErrordictgetr._already_create_accumulatorr8_is_dtype_fp16_or_bf16dtype_create_master_weight_add_accumulator_momentum_acc_str_mean_square_acc_str_mean_grad_acc_straddwarningswarn)r;blockr+pmaster_ps r"_create_accumulatorsRMSProp._create_accumulatorssx%)//399!=>>=> > j$ ' '#1JAvv999$$)D)DQWW)M)M55a8%%d&<&>=> > nd + +!55nEN33  " "N1$5 66  % %~a'8 44  # #^A%6 ++ 0K0K 1  # #1   !2!7!7 8  " # # OOq!q!%%n5   (*&q)&-) $ 5 5n E F+1-)!0, G(5}%,9()YY#}}!YY $ $  #) J r!cHURSURS5UlURSURS5UlURSURS5UlURSURS5UlURS5nU$)Nrrrrr@)rEr:r5r4r6r7)r;r+s r"rhRMSProp._update_param_groupOs"y$2D2DY2OP NN5$*<*r)FNNNN)r*zfloat | LRSchedulerrfloatrrxrrxrboolr+z;Sequence[Tensor] | Sequence[_RMSPropParameterConfig] | Noner,z%float | WeightDecayRegularizer | Noner-zGradientClipBase | Noner.z str | NonereturnNone)rrrr__doc__rKrLrMr2rTrtrhr __classcell__)r<s@r"r$r$2sxt#($  >B-11 *1 1  1  1  1  H1 <1 +1 1  1 1 f9>JX  r!r$) __future__rrOtypingrtyping_extensionsrpaddlerrbaser base.frameworkr optimizerr collections.abcr rpaddle.nn.cliprpaddle.optimizer.lrrpaddle.regularizerrrr__all__r$rr!r"rsY# )3 (-//9+$"2$ gigr!