[FIXED] How to plot timeseries from grouped logs with pandas?

Issue

I'm trying to analyze a log file using Pandas. I want to plot three lines for the count of levels "ERROR", "INFO", and "WARN" per second. With x = date (seconds), y = count.

After importing my log file, my data frame looks like this:

df_logs

I floor the date per second:

df_logs['date'] = df_logs['date'].dt.floor('S')

Then I group by message level:

ds_grouped = df_logs.groupby(['date','level'])['level'].count()

From here, I'm completely stuck:

type(ds_grouped)
> pandas.core.frame.DataFrame

I guess the correct seaborn plot is:

sns.lineplot(x='date', 
             y='count',
             hue='level', 
             data=ds_grouped)

How to plot the grouped data frame?

Solution

Here is a way to create the plot, IIUC:

# create test data
import numpy as np
import pandas as pd

n = 10_000
np.random.seed(123)
timestamps = pd.date_range(start='2020-08-27 09:00:00', 
                           periods=60*60*4, freq='1s')
level = ['info', 'info', 'info', 'warn','warn', 'error']

df = pd.DataFrame(
    {'timestamp': np.random.choice(timestamps, n), 
     'level': np.random.choice(level, n),})
print(df.head())

            timestamp  level
0 2020-08-27 09:59:42   info
1 2020-08-27 12:14:06   warn
2 2020-08-27 09:22:26   info
3 2020-08-27 12:24:12  error
4 2020-08-27 10:26:58   info

Second, sample in 5-minute intervals. You can change frequency in pd.Grouper below:

t = (df.assign(counter = 1)
     .set_index('timestamp')
     .groupby([pd.Grouper(freq='5min'), 'level']).sum()
     .squeeze()
     .unstack())
print(t.head())

level                error  info  warn
timestamp                             
2020-08-27 09:00:00     35   123    66
2020-08-27 09:05:00     32    91    73
2020-08-27 09:10:00     41   113    64
2020-08-27 09:15:00     32   110    66
2020-08-27 09:20:00     35   107    61

Third, create the plot with t.plot();

Answered By - jsmart