Optimise filtering function for runtime in pandas

Question

def filter_data(df, raw_col,threshold,filt_col):
    df['pct'] = None
    df[filt_col] = None
    df[filt_col][0] = df[raw_col][0]
    max_val = df[raw_col][0]
    for i in range(1,len(df)):
        df['pct'][i] = (df[raw_col][i] - max_val)*1.0 / max_val
        if abs(df['pct'][i]) < threshold:
            df[filt_col][i] = None
        else:
            df[filt_col][i] = df[raw_col][i]
            max_val = df[raw_col][i]
    df = df.dropna(axis=0, how='any').reset_index()
    return df


from random import randint
some_lst = [randint(50, 100) for i in range(0,50)]
some_df = pd.DataFrame({'raw_col':some_lst})
some_df_filt = filter_data(some_df,'raw_col',0.01,'raw_col_filt')

The goal to create a new column(filt_col) where record from numeric column (raw_col) are removed with the following logic; if rate of change between two adjacent rows is less than threshold remove the latter. It works but is very inefficient in terms of running time. Any hints on how I could optimise it?

Answer 1

IIUC, you can do this very simply using .pct_change() and loc

First

df['pctn'] = df.raw_col.pct_change()

Then

threshold  = 0.01
df.loc[df.pctn.abs() >= threshold]

You can check that this solution yields the same result as yours, which you said works, but is slow

df.loc[df.pctn.abs() >= 0.01].raw_col.tolist() == some_df_filt.raw_col.tolist()
True