Seaborn heatmap | How to make a heatmap in Python Seaborn and adjust the heatmap style 

Well done, superbly presented Kim

Oh wow -- so glad I could help!

Oh awesome! The more I use Seaborn, the more I appreciate all the cool features. But you're right -- the docs don't always do it justice!

Oh excellent! Very glad to hear you were spared some frustration!

Hi there 👋-- Welcome and thanks for subscribing! Glad to hear you enjoyed the video!

Oh, thank you! Very glad to hear you enjoyed it, and I will continue making other videos as well.

Why thank you! 😊

Very happy to hear this -- glad this video could help!

Most welcome! Glad to hear you liked it!

Right on! ⚡ Happy to have you subscribed and hoping you are enjoying seaborn!

So glad to hear the video and GitHub link helped!

Awww - thank you! So glad it helped and that you found what you were looking for!

You got it! And thanks very much 😀

Thanks!

Thanks so much - cheers!

Thank you! Glad you liked it 😀

Thank you - glad you enjoyed it!

Most welcome! 😄

That's a great suggestion. I almost included mask but then decided not to for timing reasons. Maybe I will make a Seaborn Q&A video in the future! For now, I do have a bit of code to show what the mask argument does as the second question here: github.com/kimfetti/Videos/blob/master/Seaborn/Q_and_A/11_heatmap_Q_and_A.ipynb Hope that helps!

Wonderful -- glad this video was useful for you!

So glad you enjoyed this tutorial. And comments like these definitely help with RU-vid rankings -- so thank YOU! 👍

Excellent - glad you enjoyed it!

Oh thank you - slowly but surely!

Excelente--de nada 😀

Good questions! Yes, one of the big uses for the correlation maps is that before building a linear regression model, say, you could see how correlated your dependent variables were with each other and potential drop variables that are highly correlated with one another or decide to use regularization. Or you might look to see which dependent variables are correlated with your independent variable to know which features to include in your regression model. Or just to do exploratory analysis and notice trends. Lots of options! Regarding your second question: yes, this is possible through the Seaborn heatmap's mask argument. The main idea of mask is that you can make some of the heatmap's rectangles blank. So you could potentially mask all the rows you didn't want with just a few lines of code, but you would then need to do additional styling like removing some of the tick labels, etc. There's an example of the mask near the bottom of this notebook if you want to check it out: github.com/kimfetti/Videos/blob/master/Seaborn/Q_and_A/11_heatmap_Q_and_A.ipynb

Very interesting question! I tested it out, and Seaborn does allow multi-indexed pandas dataframes (e.g. grouped by multiple categories). Seaborn will just hyphenate the row names; for example, if I group by origin and model year, the row names would be "europe-70" or "usa-82". You just have to make sure to input a pandas dataframe not a series, and the other thing to watch out for is scale. For example, if I look at average cylinders and average displacement, my heatmap won't show much information for average cylinders since the values are much smaller than average displacement. I added some additional code to my GitHub as a demo: github.com/kimfetti/Videos/blob/master/Seaborn/Q_and_A/11_heatmap_Q_and_A.ipynb

Oooh -- that sounds fun! I'll definitely look into that. I have also been considering doing a series about Plotly so that we can interact with the visuals. Maybe I'll see if Plotly can do this. 😄

Yes -- definitely good to know where the data is coming from. I typically use data from the Seaborn library itself for most of my videos. One of the first lines of code in this video, "cars = sns.load_dataset('mpg')" loads the car data from Seaborn and saves it in the dataframe called cars. 👍

Whoa -- you read my mind! My next video IS about saving matplotlib figures. And the same code also works for seaborn plots. Video should be out early next week 😄

@@KimberlyFessel Awesome :) Thanks

Thanks for the suggestion! I've been considering expanding out to more machine learning concepts. Hopefully at some point!

@@KimberlyFessel Yes Please!

Oh interesting! You want to overlay a heatmap on top of an picture image? I think you might have luck using matplotlib pyplot (plt): that is, plt.imshow() for the image part. This solution might help you get started: stackoverflow.com/questions/50091591/plotting-seaborn-heatmap-on-top-of-a-background-picture?rq=1 though you may need to adjust some things if you want elements at specific x- and y-coordinates.

@@KimberlyFessel Thank You

Sure thing - I usually update the size of my seaborn visuals with matplotlib. Include "from matplotlib import pyplot as plt" at the top of you code and then you can use plt.figure(figsize=(10, 8)), say, right about your heatmap code. That updates your figure to be 10 inches wide and 8 inches tall.

Yes -- good question! I almost included this option in the video but decided not to for time reasons. The mask argument lets you build heatmaps without some of the rectangles. The numpy array you pass to mask should be the same size as your heatmap dataframe/array. So if you have a = np.array([[1, 2, 3], [4, 5, 6]]) which is a 2x3 matrix, mask will also be 2x3 but will contain either True or False: True if you want to mask/remove the rectangle or False if you want the rectangle to appear. Let's say mask=np.array([[False, False, False], [True, True, True]]). Then only the top row of rectangles will show up if you do sns.heatmap(a, mask=mask). I also added your question to my Q&A doc if you want to see that code in action: github.com/kimfetti/Videos/blob/master/Seaborn/Q_and_A/11_heatmap_Q_and_A.ipynb

Hey Kimberly, it's really heartening that you replied. Thanks for that. Highly appreciated. 💛

Yes! I almost included this in my video, but the problem should be fixed in the latest version of matplotlib. This issue occurred with matplotlib 3.1.1, but you can now upgrade your version of matplotlib and it should be fixed. If you don't want to upgrade, you can change the y-limits of your heatmap. After you make your heatmap, add the following code which will decrease the plot area of the heatmap: bot, top = plt.ylim() bot += 0.5 top -= 0.5 plt.ylim(bot, top)

Yes -- this is a great suggestion! I started working on a video like this last weekend but haven't fully figured out how I want it to go yet. I will keep working on it!

oh sory i found it, its a built in datasets

I currently teach full time at a data science bootcamp called Metis, so I am not available for private training at this time. But I will continue posting more videos!