classification_crypto_transform.py

def transformData(df, days=1):
    """
    Transforming data into X variables for training.  Uses percent change and 
    multiplies the percentage by 100 rounded to 2 decimal places.
    """
    # Transforming data
    new_df = df.pct_change(
        days
    ).apply(
        lambda x: round(x*100, 2)
    ).replace(
        [np.inf, -np.inf], 
        np.nan
    ).shift(1) # Compensating for indicator lookahead bias
    
    # Dropping Nans
    new_df = new_df.dropna(
        thresh=round(new_df.shape[1]*.7) # If 70% of the values in the row are Nans, drop the whole row
    ).dropna(
        axis=1,
        thresh=round(new_df.shape[0]*.7) # If 70% of the values in the columns are Nans, drop the whole column
    )
    
    # Determining percent change and setting it as -1, 0, and 1 as Sell, Do Nothing, and Buy respectively
    new_df['decision']= df['Open'].pct_change(
        days
    ).shift(
        -days
    ).apply(
        lambda x: x>0
    ).astype(float)
    
    # Saving the last value in the dataset for later
    last_val = new_df.tail(1).drop('decision',
                                   axis=1)

    # Dropping the last NaNs from the Y variable
    new_df = new_df.dropna(
        subset=['decision']
    )

    # Filling in the rest of the NaNs with the most recent value
    new_df = new_df.fillna(method='ffill').dropna()

    return new_df, last_val