diff_transformer.py

import pandas as pd 
import numpy as np

class DiffNormalizer():

    def __init__(self, diff_len=1):
        self.diff_len = diff_len
        self.fitted = False
        self.first_values = None 

    def fit_transform(self,in_df, dropna, columns=[]):

        assert type(in_df)==pd.DataFrame
        assert set(columns).issubset(set(in_df.columns)) or columns == []

        self.first_values = in_df.iloc[0,:]
        self.fitted=True
        
        if columns:
            in_df = in_df[columns].diff(self.diff_len)
        else:
            in_df = in_df.diff(self.diff_len)

        if dropna:
            in_df.dropna(inplace=True)
            
        return in_df
        

    def inverse_transform(self, in_df, restorena, columns=[]):
        if not self.fitted:
            raise Exception('Trying to use inverse_transform but there was no fitting before!')

        if not columns:
            columns = in_df.columns

        if restorena:
            in_df = pd.concat([pd.DataFrame({k:[0] for k in columns}, columns=columns),in_df])
        for col in columns:
            print("col",col)
            cumsum = in_df[col].cumsum()
            in_df[col] = cumsum.fillna(0) + self.first_values[col]

        return in_df