Seaborn: distplot being very slow

[amelio-vazquez-reina]

I noticed that sometimes, depending on the data, KDE fitting takes a quite a bit to run (~ 3 minutes per plot). Are there any options that could limit, perhaps, the number of iterations, or still alternatively simplify the fitting process at the expense of accuracy?

[mwaskom]

Do you have statsmodels installed? I think if you do seaborn will use its FFT-based algorithm, which should be faster. Otherwise not really, although you could randomly subsample your own data.

[amelio-vazquez-reina]

Thank you @mwaskom. I have statsmodels installed, and I am actually noticing that distplot takes a long time even with rug=False, kde=False, and norm_hist=False. No idea what's to blame.

• Here is the output of pip freeze (I did this 10 minutes ago: pip install --upgrade git+git://github.com/statsmodels/statsmodels@master)
• Here is the array
• And below is the code that takes ~ 2-3 minutes to run on a modern MacBook pro with 16 GB of RAM

python f, ax = plt.subplots(figsize=(6,6)) sns.distplot(my_array, rug=False, kde=False, norm_hist=False)

More data (OS X Yosemite):

• $ uname -a:

Darwin macbook-pro.my.company.net 14.3.0 Darwin Kernel Version 14.3.0: Mon Mar 23 11:59:05 PDT 2015; root:xnu-2782.20.48~5/RELEASE_X86_64 x86_64

[mwaskom]

If you have a huge dataset, it may just be drawing a large number of bins. In that case the delay is probably just matplotlib drawing all the bars.

It's probably not that useful to draw more than ~50 bins anyway, and in 0.6.dev the automatic calculation is capped at that value to avoid this. 

—
Sent from Mailbox

On Mon, Jun 8, 2015 at 10:03 AM, Amelio Vazquez-Reina
[email protected] wrote:

Thank you @mwaskom. I have statsmodels installed, and I am actually noticing that distplot takes a long time even with rug=False, kde=False, and norm_hist=False. No idea what's to blame.

• Here is pip freeze
• Here is the array
• And below is the code:

  f, ax = plt.subplots(figsize=(6,6))
  sns.distplot(change_df.dropna(subset=['diff'])['diff'].values, rug=False, kde=False, norm_hist=False)

---

Reply to this email directly or view it on GitHub:
https://github.com/mwaskom/seaborn/issues/587#issuecomment-110074176

[amelio-vazquez-reina]

Totally correct. Thank you @mwaskom Should have thought about that!