Seaborn lineplot high cpu; very slow compared to matplotlib

Question

I have the following dataframe.

In [12]: dfFinal
Out[12]: 
           module                                            vectime                                           vecvalue
1906  client1.tcp  [1.1007512, 1.1015024, 1.1022536, 1.1030048, 1...  [0.0007512, 0.0007512, 0.0007512, 0.0007512, 0...
1912  client2.tcp  [1.10079784, 1.10159568, 1.10239352, 1.1031913...  [0.00079784, 0.00079784, 0.00079784, 0.0007978...
1918  client3.tcp  [1.10084448, 1.10168896, 1.10258008, 1.1036111...  [0.00084448, 0.00084448, 0.00089112, 0.0010310...

I want to plot the timeSeries vecvalue vs vectime for each module.

The result is the following:

To do so I can do as follows:

1) Matplotlib

start = datetime.datetime.now()

for row in dfFinal.itertuples():
    t = row.vectime
    x = row.vecvalue
    x = runningAvg(x)
    plot(t,x)

total = (datetime.datetime.now() - start).total_seconds()
print("Total time: ",total)

Doing so, takes 0.07005 seconds to accomplish.

2) Seaborn

start = datetime.datetime.now()

for row in dfFinal.itertuples():
    t = row.vectime
    x = row.vecvalue
    x = runningAvg(x)
    DF = pd.DataFrame({'x':x, 't':t})
    sns.lineplot(x='t', y='x', data=DF)

total = (datetime.datetime.now() - start).total_seconds()
print("Total time: ",total)

Doing so, takes 19.157463 seconds to accomplish.

Why is there such a huge difference? What is it that I'm doing so wrong that it takes that long to process a rather small DF?

Answer 1

Set ci=None in the call to lineplot; otherwise, confidence intervals will be computed resulting in some expensive (and unnecessary) df.groupby calls.

An aside: the snakeviz module is a great tool for quickly finding computational bottlenecks.