[New] show cluster map with new method Scenario.corr()

[lisphilar]

Summary of this new feature

To find useful indicators for ODE parameter prediction, seaborn.clstermap() will be helpful. This shows correlation coefficients and peeform clustering. Results will be shonone figure.

Ne method Scenario.corr() will show cluster map and return correlation coefficients as a dataframe.

[lisphilar]

Users will specify delay period (lag). Indicator should be delayed.

[lisphilar]

For more option, seaborn.pairplot() is also useful t check whether values are imbalanced or not with histograms.

Also, sizes (tentative) argument will be added to select the size of groups (hue). We can investigate the difference of train/test data.

If this is implement in RegressionHandler, argument names may be different.

[lisphilar]

seaborn-analyzer package could be used.
https://github.com/c60evaporator/seaborn-analyzer

[lisphilar]

To detect non-linear correlation, minepy package can be used.
https://minepy.readthedocs.io/en/latest/

New method should detect non-linear correlation when user need to do (linear=False).

[lisphilar]

Memo for visualization (seaborn)
https://link.medium.com/lolY32qDDib

[lisphilar]

Memo regarding coding.

X = snl.records(variables="all", show_figure=False)
X = X.drop(["Susceptible", "Confirmed", "Infected", "Recovered", "Fatal", "Date"], axis=1)
X = X.apply(lambda x: (x-x.mean())/ x.std(), axis=0).dropna(axis=1)
Y = snl.track()
Y = Y.loc[Y.Scenario == "Main", ["theta", "kappa", "rho", "sigma"]]
df = pd.concat([Y, X], axis=1)
df.corr().style.background_gradient(cmap="coolwarm")

[lisphilar]

Highlight corr and PCA.
https://gist.github.com/lisphilar/ac0c0c97d780335c03262d29f2b1e5aa

[lisphilar]

Or we can do

import seaborn as sns
df = pd.concat([Y, X], axis=1)
df = df.corr()
mask = np.zeros_like(df)
mask[np.triu_indices_from(mask)] = True
sns.heatmap(df, mask=mask, square=True);

[lisphilar]

Cliuster map:

df = pd.concat([Y, X], axis=1)
df = df.corr()
sns.clustermap(df, square=True);