def _plot_perm_imp(perm, test_sample, node_params, **kwargs):
weights = dict(
zip(test_sample.columns.tolist(), perm.feature_importances_))
if node_params is None:
node_params = {}
node_flg = True
else:
node_flg = False
node_weights = {}
for node, val in weights.items():
if len(node) > 1:
continue
if node_flg:
node_params.update({
node[0]:
'nice_node' if weights[node] >= 0 else 'bad_node',
})
node_weights.update({node[0]: val})
edge_cols = [i for i in test_sample.columns if len(i) == 2]
if len(edge_cols) == 0:
print(
"Sorry, you use only unigrams, change ngram_range to (1, 2) or greater"
)
return
data = []
for key in edge_cols:
data.append([key[0], key[1], weights.get(key)])
plot.graph(pd.DataFrame(data),
node_params,
node_weights=node_weights,
**kwargs)
def plot_graph(self, user_based=True, node_params=None, **kwargs):
"""
Create interactive graph visualization
:param user_based: if True, then edge weights is calculated as unique rate of users who go through them
:param node_params: mapping describes which node should be highlighted by target or source type
Node param should be represented in the following form
```{
'lost': 'bad_target',
'passed': 'nice_target',
'onboarding_welcome_screen': 'source',
}```
If mapping is not given, it will be constracted from config
:param kwargs: other parameters for visualization
:return: Nothing
"""
if user_based:
kwargs.update({
'edge_col': self.retention_config['index_col'],
'edge_attributes': '_nunique',
'norm': True,
})
if node_params is None:
_node_params = {
'positive_target_event': 'nice_target',
'negative_target_event': 'bad_target',
'source_event': 'source',
}
node_params = {}
for key, val in _node_params.items():
name = self.retention_config.get(key)
if name is None:
continue
node_params.update({name: val})
plot.graph(self._obj.trajectory.get_edgelist(**kwargs), node_params, **kwargs)