#!/usr/bin/env python
# coding: utf-8

# # Assessing feature performance
# 
# DWPC = 0.4, computed on a [subset](http://doi.org/10.15363/thinklab.d115#1) of non-indications

# In[1]:


import re

import matplotlib.pyplot
import pandas
import mpld3
import mpld3.plugins
import seaborn

get_ipython().run_line_magic('matplotlib', 'inline')


# In[3]:


auc_df = pandas.read_table('data/auc.tsv')
# Add metapath length
auc_df['length'] = auc_df.metapath.map(lambda x: len(re.split(r'[a-z<>]+', x)) - 1)
auc_df.head(2)


# ## Performance by path length

# In[4]:


ax = seaborn.violinplot(x='auroc', y='length', data=auc_df, cut=0, orient='h', inner='quartile')


# In[5]:


ax = seaborn.stripplot(x='auroc', y='length', data=auc_df, jitter=0.25, orient='h')


# ## Feature AUROC versus non-zero fraction

# In[6]:


grid = seaborn.jointplot('nonzero', 'auroc', data=auc_df, joint_kws={'alpha': 0.5})
points = grid.ax_joint.collections[0]
labels = auc_df.metapath.tolist()
tooltip = mpld3.plugins.PointLabelTooltip(points, labels)
mpld3.plugins.connect(grid.fig, tooltip)
mpld3.display()


# ## Feature AUPRC versus AUROC

# In[7]:


grid = seaborn.jointplot('auroc', 'auprc', data=auc_df, joint_kws={'alpha': 0.5})
points = grid.ax_joint.collections[0]
labels = auc_df.apply(lambda x: '{}: {:.1f}% nonzero'.format(x['metapath'], x['nonzero']*100), axis=1).tolist()
tooltip = mpld3.plugins.PointLabelTooltip(points, labels)
mpld3.plugins.connect(grid.fig, tooltip)
mpld3.display()