patternnet with tall arrays

1 次查看（过去 30 天）

Giorgio De Nunzio 2020-4-25

2
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/520682-patternnet-with-tall-arrays

Hi all,

I've begun studying tall arrays for my big-data applications (binary classification of medical data, with a multilayer perceptron). First I put all my feature data files (describing my samples) into a datastore, and then I used a tall array to represent the data, to later submit them to a shallow ANN for supervised classification.

By cumulating data from many patiens, I've built a very tall feature matrix where rows are the (many, many...) samples, and columns are the different features.

In the past I've used normal memory arrays (structured like the matrices described above, where rows are the samples, and columns are the features) that were transposed before feeding the train function of a patternnet network. The reason of transposing was of course that patternnet expects a matrix where rows are the features and each column is a sample.

Now, unfortunately I cannot transpose a tall array, and I cannot gather and then transpose it for memory rasons....

My question is: how can I use a matrix like the one described before (each row representing a sample and each column a feature), but contained in a tall array, with patternet? Can I? Should I avoid patternet and use other approaches and functions? I think that patternnet (and its relatives) is the only classsification function in Matlab which does not want the more or less standard "samples in rows, features in columns" structure....

Thanks

Giorgio