• The supported formats are: 

  • CSV with ";" or "," separator; 
  
  • NumPy .npy
  
  • MATLAB MAT-files .mat

• Files contain the numerical data, allowed types: int, float, double

• Any **input file** contains a table with:

  • number of lines equal to a number of samples
   
  • number of columns equal to a number of input features
   
  • CSV files could have one additional line for a header
   
• Any **output file** contains a table with:

  • number of lines equal to a number of samples
   
  • number of columns equal to a number of output features, for Classification these features are the classes
   
  • the outputs are one-hot encoded: each line contains '0' on all positions, except for one containing '1'. This position corresponds to a class to which belongs the sample on the line.
  
  • CSV files could have one additional line for a header
   
• **input file** and the corresponding **output file** have the same number of samples

• The data can be provided in chunks, in multiple **input** and **output files**. In this case pay attention to preserving the correspondence between **input** and **output files**

There is no need to normalize the data, as the normalization is handled by NeurEco, :std:ref:`Normalizing the data`.
  