Follow some plots of the data from rounds 300 to 332…
First we plot the well known problem of TC being very poorly correlated with other metrics, such as model correlation and fncV3.
Model TC vs CORR
Model TC vs FncV3
Same information but shown by round:
Model TC vs CORR by Round
Model TC vs FncV3 by Round
Now I want to see if the relationship of TC with CORR and FNCV3 is somehow influenced by “Model Correlation with Meta Model” or stake amount. So i split the “Model Correlation with Meta Model” and the stake amount in 9 bins.
Model TC vs CORR by Correlation with Meta Model
Model TC vs FNCV3 by Correlation with Meta Model
Model TC vs CORR by Stake
Model TC vs FNCV3 by Stake
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
df = pd.read_csv('rounds-300-332.csv')
sns.jointplot(data=df, x='corr', y='tc', kind="reg", truncate=False)
sns.jointplot(data=df, x='fncV3', y='tc', kind="reg", truncate=False)
plt.show()
TCvsCORR = df.groupby(['roundNumber']).apply(lambda x: x.tc.corr(x['corr']))
TCvsCORR.name='PearsonCoeff(TC,CORR)'
pd.DataFrame(TCvsCORR).reset_index().plot(x='roundNumber',y='PearsonCoeff(TC,CORR)',kind='line')
TCvsFNCV3 = df.groupby(['roundNumber']).apply(lambda x: x.tc.corr(x['fncV3']))
TCvsFNCV3.name='PearsonCoeff(TC,FNCV3)'
pd.DataFrame(TCvsFNCV3).reset_index().plot(x='roundNumber',y='PearsonCoeff(TC,FNCV3)',kind='line')
plt.show()
df['corrWMetamodelBin'] = pd.cut(df['corrWMetamodel'], 9, labels=False)
df['stakeBin'] = pd.qcut(df['selectedStakeValue'].rank(method='first'), 9, labels=False)
sns.lmplot(data=df, x='corr', y='tc', col='stakeBin', col_wrap=3, truncate=False, scatter_kws={"alpha": 0.6})
sns.lmplot(data=df, x='fncV3', y='tc', col='stakeBin', col_wrap=3, truncate=False, scatter_kws={"alpha": 0.6})
sns.lmplot(data=df, x='corr', y='tc', col='corrWMetamodelBin', col_wrap=3, truncate=False, scatter_kws={"alpha": 0.6})
sns.lmplot(data=df, x='fncV3', y='tc', col='corrWMetamodelBin', col_wrap=3, truncate=False, scatter_kws={"alpha": 0.6})
plt.show()