NASA Ocean Color

ocssw V2022

 from ._CustomTransformer import _CustomTransformer
  
 import numpy as np 
  
  
 class BaggingColumnTransformer(_CustomTransformer):
     ''' 
     Randomly select a percentage of columns to drop, always keeping the
     band features (first n_bands columns). Optionally, the last n_extra 
     columns will also be always kept as additional features.
     '''
  
     def __init__(self, n_bands, *args, n_extra=0, percent=0.75, seed=None, **kwargs):
         self.n_bands = n_bands
         self.n_extra = n_extra
         self.percent = percent
         self.random  = np.random.RandomState(seed)
  
     def _fit(self, X, *args, **kwargs):
         shp  = X.shape[1] - self.n_bands
         ncol = int(shp*self.percent)
         cols = np.arange(shp-self.n_extra) + self.n_bands
         self.random.shuffle(cols)
         new_cols  = list(cols[:ncol])
         if self.n_extra: new_cols += list(X.shape[1]-(np.arange(self.n_extra)+1))
         self.cols = np.append(np.arange(self.n_bands), new_cols, 0)
         # print(f'Reducing bands from {shp} ({X.shape[1]} total) to {ncol} ({len(self.cols)} total) ({self.cols})')
  
     def _transform(self, X, *args, **kwargs):
         return X[:, self.cols.astype(int)]