Thompson Sampling in Python

We have provided a detailed example of Thompson Sampling using R. Below, we show how we can run the Multi-armed bandit algorithm in Python.

import pandas as pd
import numpy as np
from scipy.stats import beta
import math

np.random.seed(1234)

def bandits(data, beta_values=1000):
    indexes=data.index
    data['random_state']=range(0,len(data))
    data['monte_carlo']=data.apply(lambda x:beta(x['clicks']+1,x['impressions']-x['clicks']+1).rvs(beta_values),axis=1)
    weights=pd.DataFrame(list(data['monte_carlo']),index=indexes).idxmax().value_counts()/beta_values
    data['weights']=weights
    return(data[['weights']].fillna(0))
	
	
### example

z={'clicks': [15,17,19],
 'impressions': [1000,1000,1000]}
z=pd.DataFrame(z)

# and the weights
print(bandits(z))

Output:

	weights
0	0.159
1	0.288
2	0.553