Error in helperModClassTrainingOptions (line 29) 'CheckpointPath',checkpointPath,...

Question

john karli 2022-2-16

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1651220-error-in-helpermodclasstrainingoptions-line-29-checkpointpath-checkpointpath

评论： Joss Knight 2022-2-18

采纳的回答： Joss Knight

在 MATLAB Online 中打开

I want to train the model using following link

https://www.mathworks.com/help/deeplearning/ug/modulation-classification-with-deep-learning.html

I want to save every epochs but when i run the following section

checkpointPath = pwd;
maxEpochs = 20;
miniBatchSize = 128;
options = helperModClassTrainingOptions(maxEpochs,miniBatchSize,...
  numel(rxTrainLabels),rxValidFrames,rxValidLabels);
trainedNettime = trainNetwork(rxTrainFrames,rxTrainLabels,lgraph_1 ,options);
save trainedNettime

I got the error

Unrecognized function or variable 'checkpointPath'.
Error in helperModClassTrainingOptions (line 29)
'CheckpointPath',checkpointPath,...

my helperModClassTrainingOptions function is

function options = helperModClassTrainingOptions(maxEpochs,miniBatchSize,...
  trainingSize,rxValidFrames,rxValidLabels)
%helperModClassTrainingOptions Modulation classification training options
%   OPT = helperModClassTrainingOptions(MAXE,MINIBATCH,NTRAIN,Y,YLABEL)
%   returns the training options, OPT, for the modulation classification
%   CNN, where MAXE is the maximum number of epochs, MINIBATCH is the mini
%   batch size, NTRAIN is the number of training frames, Y is the
%   validation frames and YLABEL is the labels.
%
%   This function configures the training options to use an SGDM solver.
%   By default, the 'ExecutionEnvironment' property is set to 'auto', where
%   the trainNetwork function uses a GPU if one is available or uses the
%   CPU, if not. To use the GPU, you must have a Parallel Computing Toolbox
%   license. Set the initial learning rate to 2e-2. Reduce the learning
%   rate by a factor of 10 every 9 epochs. Set 'Plots' to
%   'training-progress' to plot the training progress.
%   
%   See also ModulationClassificationWithDeepLearningExample.
%   Copyright 2019 The MathWorks, Inc.
validationFrequency = floor(trainingSize/miniBatchSize);
options = trainingOptions('sgdm', ...
  'InitialLearnRate',1e-3, ...
  'MaxEpochs',maxEpochs, ...
  'MiniBatchSize',miniBatchSize, ...
  'Shuffle','every-epoch', ...
  'Plots','training-progress', ...
  'CheckpointPath',checkpointPath,...
  'ValidationData',{rxValidFrames,rxValidLabels}, ...
  'ValidationFrequency',validationFrequency, ...
  'Verbose',false, ...
   'LearnRateSchedule', 'piecewise', ...
  'LearnRateDropPeriod', 9, ...
  'LearnRateDropFactor', 0.1);
  

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Joss Knight 2022-2-16

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1651220-error-in-helpermodclasstrainingoptions-line-29-checkpointpath-checkpointpath#answer_897945

You need to pass the checkpointPath variable to your function.

5 个评论
显示 3更早的评论隐藏 3更早的评论

yanqi liu 2022-2-17

在 MATLAB Online 中打开

function options = helperModClassTrainingOptions(maxEpochs,miniBatchSize,...
    trainingSize,rxValidFrames,rxValidLabels)
%helperModClassTrainingOptions Modulation classification training options
%   OPT = helperModClassTrainingOptions(MAXE,MINIBATCH,NTRAIN,Y,YLABEL)
%   returns the training options, OPT, for the modulation classification
%   CNN, where MAXE is the maximum number of epochs, MINIBATCH is the mini
%   batch size, NTRAIN is the number of training frames, Y is the
%   validation frames and YLABEL is the labels.
%
%   This function configures the training options to use an SGDM solver.
%   By default, the 'ExecutionEnvironment' property is set to 'auto', where
%   the trainNetwork function uses a GPU if one is available or uses the
%   CPU, if not. To use the GPU, you must have a Parallel Computing Toolbox
%   license. Set the initial learning rate to 2e-2. Reduce the learning
%   rate by a factor of 10 every 9 epochs. Set 'Plots' to
%   'training-progress' to plot the training progress.
%
%   See also ModulationClassificationWithDeepLearningExample.
%   Copyright 2019 The MathWorks, Inc.
if ~exist('checkpointPath','var')
    checkpointPath = fullfile(pwd, 'tmp');
end
if ~exist(checkpointPath, 'dir')
    mkdir(checkpointPath);
end
validationFrequency = floor(trainingSize/miniBatchSize);
options = trainingOptions('sgdm', ...
    'InitialLearnRate',1e-3, ...
    'MaxEpochs',maxEpochs, ...
    'MiniBatchSize',miniBatchSize, ...
    'Shuffle','every-epoch', ...
    'Plots','training-progress', ...
    'CheckpointPath',checkpointPath,...
    'ValidationData',{rxValidFrames,rxValidLabels}, ...
    'ValidationFrequency',validationFrequency, ...
    'Verbose',false, ...
    'LearnRateSchedule', 'piecewise', ...
    'LearnRateDropPeriod', 9, ...
    'LearnRateDropFactor', 0.1);

john karli 2022-2-18

training issue.PNG

why my validation get sudden rise in last epochs. I have attached the image.

Joss Knight 2022-2-18

The final validation is computed after a final epoch to compute the batch normalization statistics. Some networks are particularly sensitive to the difference between the mini-batch statistics and those of the whole dataset. Make sure your dataset is shuffled and your minibatch size is as large as possible. To avoid this (at a small additional performance cost), using moving averages (see BatchNormalizationStatistics training option).

I can't explain why it's not checkpointing the network every epoch.

请先登录，再进行评论。