Ufdjrw

I have a time-series related question on how to plot time stamps along a horizontal axis for multiple cases attributed to the same person. Let me explain:

Let us assume we have Jason and Georgia. Both of them work on different cases, which largely have these potential "events": start, pause, resume, end. Many cases just have a "start" and an "end", whereas other also include a pause-resume interval. While one case is paused, the user can work on a different case. I have all this information on a Pandas DataFrame, and I collect the user-and-case-level information doing a groupby.

Sample data for reproducible code (assuming pandas and numpy are imported):

raw_data = 'user': ['Jason', 'Georgia', 'Jason', 'Jason', 'Georgia'], 
 'case': ['a', 'b', 'c', 'd', 'e'], 
 'date_picked_up': ['2018-10-25 14:06', '2019-01-25 10:44', '2019-01-25 09:14', '2019-01-25 12:12', '2019-02-21 10:01'], 
 'date_paused': ['2018-10-26 11:08', '2019-01-25 12:11', np.nan, np.nan, '2019-02-21 12:37'],
 'date_resumed': ['2018-10-26 11:20', '2019-01-25 15:21', np.nan, np.nan, '2019-02-21 13:24'],
 'date_closed': ['2018-10-29 16:57', '2019-01-25 16:34', '2019-01-25 11:46', '2019-01-25 15:24', '2019-01-25 13:56']
df = pd.DataFrame(raw_data, columns = ['user', 'case', 'date_picked_up', 'date_paused', 'date_resumed', 'date_closed'])
df

This returns df, a pandas DataFrame with the progression of each case. When we do not have the pause-resume interval, the values are np.nan. Pandas groupby automatically ignores nan values, which we do not want, so in order to deal with this I am using fillna with a Timestamp in 1900, after applying to all columns pd.to_datetime:

date_cols = ['date_picked_up', 'date_paused', 'date_resumed', 'date_closed']
for c in date_cols:
 df[c] = pd.to_datetime(df[c], format='%Y%m%d %H:%M')

Now the best way I found to aggregate the data by user and then by case is:

df.fillna(pd.Timestamp('19000101'))
 .groupby(['user', 'case', 'date_picked_up', 'date_paused', 'date_resumed', 'date_closed'])[['date_picked_up', 'date_paused', 'date_resumed', 'date_closed']].count()

My goal (from this sample data) is two plots, one for Jason and one for Georgia, where the timestamps (ideally not the 1900 ones) will show up along horizontal "lines", one for each case (on the y axis). The closest example is here: Plotting labled time series in pandas where instead of dogs cats and cows we would have (for Jason) cases a, c, and d on the y-axis.

I have found ideas on how to move everything to bokeh or d3 for what I really want (e.g.: https://github.com/jiahuang/d3-timeline, How to plot stacked event duration (Gantt Charts) using Python Pandas?), but I hope to find a solution in Python and Matplotlib/Seaborn, since I believe that my data structure is already in a good enough format.