Handling no-data points

ainazHjm / Baselines

This repo contains the baseline models for image data (hdf5 format) and a simple linear model for sea2sky data (csv format)

0 stars 0 forks source link

Handling no-data points #3

Open ainazHjm opened 5 years ago

ainazHjm commented 5 years ago

[x] replace no-data points with 0s
[ ] use the most frequent label (categorical data)
[ ] add a new feature called, other value or I don't know, for each type of feature (water, slope, surficial material, ...)

ainazHjm commented 5 years ago

Replacing no-data/NULL points with 0s is a dangerous thing because the model sees those points as does not exist not that I don't know whether I have that feature or not but the model's performance is really good. I'm using a simple linear layer for this task; I guess the reason that it's working well is that the no-data features are not really that important for prediction and we have all the necessary features.