CODEWITHSUNDEEP
javamachine-learningneural-networkdeep-learning
Slayer0248
Slayer0248

Reputation: 1261

Deeplearning4j Neural net configuration

I'm trying to get deeplearning4j to run on a dataset I'm reading in with CSVRecordReader. I'm getting one of the following errors when I run model.fit() on the dataset.

IllegalStateException: Column of left array # != rows of right # or rows of 
left array # != columns of right #

or 

IllegalStateException: Mis matched shapes

The first seems to go away when I have the number of inputs to the first layer and outputs to the second layer equal to the dimensions of the training dataset, but then I start getting the second. Do the first layer's inputs & outputs always have to be determined by the input dimensions? I'm not sure what to do about the second error message.

I'm just trying to get the neural net to fit another dataset at this point, so I'm less concerned about what the layers are doing for the moment. If it could be more that just input/output parameters causing this problem, I'll post the current neural net configuration I'm using.

Upvotes: 3

Views: 220

Answers (1)

Adam Gibson
Adam Gibson

Reputation: 3205

For future reference on how to build neural nets, please see: http://deeplearningbook.org/

I would also recommend andrew ng's intro to machine learning on coursera: https://www.coursera.org/learn/machine-learning

Finally getting to your answer here: A feature vector is a fundamental machine learning concept.

Your inputs always need to match the number of features on your feature vector.

Upvotes: 2

Related Questions