xuyxu
commented
2 years ago Hi, @JohnTaco93, this version of deep forest focuses on tabular data only. For time series data, one possible solution is to first use feature extraction techniques (e.g., tsfresh) to transform your data into a non-ordinal dataset ([1643x1250x3] -> [1643, N], N the number of features generated by tsfresh).
I am working with multivariate time series classification and would like to work with the raw data as the input of the deep forest model. I think it should be similar to the 1D CNN in which the input, in my case, is a matrix of [1643x1250x3]. This means that I have 1643 samples and each sample is a matrix of 1250x3. I want to do the same with deep forest but when I use the cascade classifier I get the following error: too many values to unpack (expected 2). When the input is a simple vector, it works, like in normal machine learning problems for tabular data. I think there should be a way to do this task since in the previous documentation (gcforest) there is an example with images in which the input is a 2D matrix (mnist dataset). Please let me know if you have an example of this case with Deep Forest. Thanks!