perf_overall.py

import pandas as pd
import matplotlib.pyplot as plt
import perfplot

def cs(df):
    df = df.copy()
    m = df.value.eq(1)  # Cached for efficiency.
    df['previous_smaller_index'] = np.where(
        m, -1, len(df) - np.triu(df.value.values < df.value[:,None]).argmax(1) - 1
    )[::-1]

    df['previous_1_index'] = (df.assign(X=m.cumsum())
                                .groupby('X')['X']
                                .transform('idxmax')
                                .mask(m, -1)) 
    return df

def cs2(df):
    df = df.copy()
    m = df.value.eq(1)  # Cached for efficiency.
    df['previous_smaller_index'] = np.where(
        m, -1, len(df) - np.triu(df.value.values < df.value[:,None]).argmax(1) - 1
    )[::-1]

    df['previous_1_index'] = (np.where(m, -1, 
        df.index.where(m).to_series(index=df.index).ffill(downcast='infer')))

    return df

def spghttCd(df):
    df = df.copy()
    df['previous_smaller_index'] = (
        df.apply(lambda x: df.index[:x.name][(x.value>df.value)[:x.name]].max(), 
                 axis=1))

    df['previous_1_index'] = (
        pd.Series(df.index.where(df.value==1)).shift()[df.value!=1].ffill())

    return df.fillna(-1, downcast='infer')

def wb(df):
    df = df.copy()
    l=list(zip(df['index'],df.value))[::-1]
    t=[]
    n=len(l)
    for x in l:
        if x[1]==1:
            t.append(-1)
        else:
            t.append(next(y for y in l[n-x[0]:] if y[1]<x[1])[0])
    df['previous_smaller_index']=t[::-1]
    df['previous_1_index']=df['index'].where(df.value==1).ffill().where(df.value!=1,-1).astype(int)

    return df

def quick_concat(df_, n):
    df = pd.concat([df_] * n, ignore_index=True)
    df['index'] = df.index
    return df

data = {
    'index': {0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6},
    'value': {0: 1, 1: 1, 2: 2, 3: 3, 4: 2, 5: 1, 6: 1}
}
df_ = pd.DataFrame(data).assign(
    previous_smaller_index=np.nan, previous_1_index=np.nan)
df_

#    index  value  prev_smaller_idx  prev_1_idx
# 0      0      1               NaN         NaN
# 1      1      1               NaN         NaN
# 2      2      2               NaN         NaN
# 3      3      3               NaN         NaN
# 4      4      2               NaN         NaN
# 5      5      1               NaN         NaN
# 6      6      1               NaN         NaN
    
perfplot.show(
    setup=lambda n: quick_concat(df_, n),
    kernels=[cs, cs2, spghttCd, wb],
    labels=['coldspeed', 'coldspeed2', 'SpghttCd', 'W-B'],
    n_range=[2**k for k in range(0, 7)],
    xlabel='N (scale for `df_`)',
    logy=True,
)