In [1]:
%matplotlib inline
import pandas as pd
import random
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
In [33]:
df = pd.read_csv('all_results.csv')
In [3]:
df.head()
Out[3]:
In [4]:
df['machine'].unique()
Out[4]:
In [5]:
df['benchmark'].unique()
Out[5]:
In [18]:
df['limits'].unique()
Out[18]:
In [34]:
df.fillna(True, inplace=True)
In [35]:
%run normalize.py
In [36]:
df = obtain_speedup(df, 'issdm-12')
In [49]:
df.head()
Out[49]:
In [71]:
machine = '192.168.140.81'
subdf = df[df["machine"] == machine]
In [72]:
sns.boxplot(subdf.speedup, groupby=subdf.limits)
Out[72]:
The above might look like it shows that with limits we get lower/higher variability, but we need to normalize speedups w.r.t. the max/min values of each 'limits' group, otherwise this is an optical illusion
In [73]:
subdf['speedup_normalized'] = subdf.apply(
lambda x:
(x['result'] - subdf[subdf['limits'] == x['limits']]['speedup'].mean()) /
(subdf[subdf['limits'] == x['limits']]['speedup'].max() -
subdf[subdf['limits'] == x['limits']]['speedup'].min()),
axis=1
)
In [77]:
subdf.head()
Out[77]:
In [75]:
sns.boxplot(subdf.speedup_normalized, groupby=subdf.limits)
Out[75]:
In [76]:
subdf[subdf['speedup'] > 25]
Out[76]:
In [ ]: