Sampling from a concatenated dataset

vision
1

I have created a concatenated dataset using `torch.utils.data.ConcatDataset, now I would like to load the data using data loader and use the sample feature. The sampler I am using is the following:

2

What kind of Dataset are you passing to ConcatDataset?
Currently the implementation you’ve linked seems to only work with the MNIST dataset and ImageFolder, see this line of code.
You could try to implement the _get_label method for your dataset or alternatively you could try to use the built-in WeightedRandomSampler.

3

Here is the code I am using :

trainset=ConcatDataset([(datasets.ImageFolder(train_dir, train_transform)),datasets.ImageFolder(train_dir, valid_transform)])                                
trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size,shuffle=True)

what I would like to do is:

trainset=ConcatDataset([(datasets.ImageFolder(train_dir, train_transform)),datasets.ImageFolder(train_dir, valid_transform)])                                
trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size,sampler=any_sampler_function(trainset),shuffle=False)

I think I tried WeightedRandomSampler but the error I get is the same as using the custom sampler, something like “a tuple was provided while a sampler object is required”.