The original code for the paper "How to train your MAML" along with a replication of the original "Model Agnostic Meta Learning" (MAML) paper in Pytorch.
Hi, AntreasAntoniou, thanks for your great job. But I have some confusion when reading the code.
In LSLRGradientDescentLearningRule Module, it need the parameter called use_learnable_learning_rates, but when instantiate the Module in few_shot_learning_system.py/MAMLFewShotClassifier, the parameter use_learnable_learning_rates is passed by self.args.learnable_per_layer_per_step_inner_loop_learning_rate which is not defined.
So I want to know how it works. The parameter use_learnable_learning_rates is used in torch.nn.Parameter(data, requires_grad=use_learnable_learning_rates), which can not be None.
If I need to define the learnable_per_layer_per_step_inner_loop_learning_rate, it is a boolean value or a list of boolean value?
Hi, AntreasAntoniou, thanks for your great job. But I have some confusion when reading the code.
In
LSLRGradientDescentLearningRule
Module, it need the parameter calleduse_learnable_learning_rates
, but when instantiate the Module infew_shot_learning_system.py/MAMLFewShotClassifier
, the parameteruse_learnable_learning_rates
is passed byself.args.learnable_per_layer_per_step_inner_loop_learning_rate
which is not defined.So I want to know how it works. The parameter
use_learnable_learning_rates
is used intorch.nn.Parameter(data, requires_grad=use_learnable_learning_rates)
, which can not beNone
.If I need to define the
learnable_per_layer_per_step_inner_loop_learning_rate
, it is aboolean value
or alist of boolean value
?