def get_step_matrix(self, max_steps=None, plot_type=True, **kwargs):
"""
Plots heatmap with distribution of events over event steps (ordering in the session by event time)
:param max_steps: maximum number of steps to show
:param plot_type: if True, then plot in interactive session (jupyter notebook)
:return: pd.DataFrame with distribution of events over event order
"""
target_event_list = self.retention_config['target_event_list']
# TODO give filter, return to desc tables ???
self._add_event_rank(**kwargs)
agg = self.get_edgelist(
cols=['event_rank', self.retention_config['event_col']],
norm=False)
if max_steps:
agg = agg[agg.event_rank <= max_steps]
agg.columns = ['event_rank', 'event_name', 'freq']
tot_cnt = agg[agg['event_rank'] == 1].freq.sum()
agg['freq'] = agg['freq'] / tot_cnt
piv = agg.pivot(index='event_name',
columns='event_rank',
values='freq').fillna(0)
piv.columns.name = None
piv.index.name = None
for i in target_event_list:
piv = piv.append(self._add_accums(piv, i))
piv = piv.round(2)
if plot_type:
plot.step_matrix(piv)
return piv
def get_step_matrix_difference(self, groups, plot_type=True, max_steps=30, **kwargs):
"""
Plots heatmap with difference of events distributions over steps between two given groups
:param groups: boolean vector that splits data in to groups
:param plot_type: if True, then heatmap plot will be shown in interactive mode
:param max_steps: maximum number of steps to show
:return: pd.DataFrame with step matrix
"""
desc_old = self._obj[~groups].copy().trajectory.get_step_matrix(plot_type=False, max_steps=max_steps, **kwargs)
desc_new = self._obj[groups].copy().trajectory.get_step_matrix(plot_type=False, max_steps=max_steps, **kwargs)
desc_old, desc_new = self._create_diff_index(desc_old, desc_new)
desc_old, desc_new = self._diff_step_allign(desc_old, desc_new)
diff = desc_new - desc_old
diff = diff.sort_index(axis=1)
if plot_type:
plot.step_matrix(diff)
return diff
def get_step_matrix(self,
max_steps=None,
plot_type=True,
dt_means=False,
**kwargs):
"""
Plots heatmap with distribution of events over event steps (ordering in the session by event time)
:param max_steps: maximum number of steps to show
:param plot_type: if True, then plot in interactive session (jupyter notebook)
:return: pd.DataFrame with distribution of events over event order
"""
target_event_list = self.retention_config['target_event_list']
# TODO give filter, return to desc tables ???
self._add_event_rank(**kwargs)
agg = self.get_edgelist(
cols=['event_rank', self.retention_config['event_col']],
norm=False)
if max_steps:
agg = agg[agg.event_rank <= max_steps]
agg.columns = ['event_rank', 'event_name', 'freq']
tot_cnt = agg[agg['event_rank'] == 1].freq.sum()
agg['freq'] = agg['freq'] / tot_cnt
piv = agg.pivot(index='event_name',
columns='event_rank',
values='freq').fillna(0)
piv.columns.name = None
piv.index.name = None
for i in target_event_list:
piv = piv.append(self._add_accums(piv, i))
piv = piv.round(2)
if plot_type:
plot.step_matrix(piv)
if dt_means:
means = np.array(
self._obj.groupby('event_rank').apply(lambda x: np.exp(
np.mean(
np.log((x.next_timestamp - x.event_timestamp).dt.
total_seconds() + 1e-20)))))
piv = pd.concat([
piv,
pd.DataFrame([means], columns=piv.columns, index=['dt_mean'])
])
return piv