How do I use datastores for sequence to sequence classification using the Deep Learning Toolbox?

1 次查看（过去 30 天）
Jason Tracey
Jason Tracey 2020-4-8
I'm using the Deep Learning Toolbox to classify input signals, in a similar fashion to that shown in this example. However instead of 3 features, my data has 152, and instead of 6 observations, my data has 767. This amount of data is too large to feed directly into network training function, and I believe datastores are the solution for this. However I am having trouble working out the correct format for the data output of the datastore read function.
The MATLAB examples for read data do not seem suited to sequence-to-sequence classification. This example shows general sequence classification, but requires padding of sequences with zeros, resulting in uniform sized sequences, which doesn't seem appropriate for sequence to sequence classification.
I have tried modifying the datastore in the previous example such that its read function outputs a cell of size {batchLength,2} with the first column containing doubles in the format {numFeatures,sequenceLength} and the second column containing categoricals in the format {1,sequenceLength}. Separating this output into its consituent columns and feeding them directly into the trainNetwork function works perfectly. However, in the datastore format I get an error:
% Error using trainNetwork (line 170)
% Unexpected response size: If the network outputs sequences, then the responses must be arrays with
% feature dimension 9.
Where 9 is the number of classes.

请先登录，再回答此问题。

回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息

标签

产品


版本

R2019b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by