Solving real world data science problems with Python! (computer vision edition)

Practice your Python Pandas data science skills with problems on StrataScratch!

In this video we work on a real world computer vision problem using Python. The problem task is to create a model that can distinguish a flower known as “La Eterna” from other types of flowers.

To do this we create convolutional neural networks (CNNs) using the Tensorflow/Keras libraries. We examine how to create a simple model and then improve it using techniques such as data augmentation & preprocessing. We play around with different types of network architectures and see how changes improve or decrease overall task performance.

Link to source code (Github):

Link to HP challenge:

My previous videos on neural networks!

*** I've left a bunch of additional useful resources in the README of the Github repo ***

Videography for clips I integrated at the start by Ryan Cabana

Hopefully you enjoy this video! Please leave it a like & subscribe if you did :).

If you have questions about topics covered in this video, please let me know in the comments.

-------------------------
Follow me on social media!
-------------------------
Song at the end
Creative Commons — Attribution 3.0 Unported — CC BY 3.0

-------------------------

Practice your Python Pandas data science skills with problems on StrataScratch!

Join the Python Army to get access to perks!

*I use affiliate links on the products that I recommend. I may earn a purchase commission or a referral bonus from the usage of these links.

-------------------------
Video timeline!
0:00 - Intro
0:40 - Video overview (what we’ll be working on)
1:53 - Code setup (GitHub repo & HP challenge link)
5:11 - Exploring the dataset that we’ll be using
8:53 - Installing necessary Python libraries (opencv-python, tensorflow)
10:31 - Reviewing template code (part 2)
11:03 - How we load in the dataset (ImageDataGenerator, flow_from_directory)
14:33 - Building our first classifier (convolutional neural net - CNN)
25:19 - Methods to improve neural network performance (MaxPooling, dropout, network architecture)
29:30 - Quick discussion about importance of precision & recall versus accuracy
32:35 - Data augmentation & preprocessing (another way to improve performance)
47:15 - Programmatically finding the best neural network architectures (Keras Tuner)
1:20:00 - Video recap & conclusion

[KeithGalli]

Sorry for the delay in getting this video posted! I think it's a good one though so hopefully it's worth the wait :)

[ilan_kayesar]

Hey man, haven't seen your channel in a while, just so you know back when I was like in the 5th grade I would watch your python tutorials which taught me how to code, this year I'm in the 9th grade and have been accepted into the best academy for code in my country, you have made an incredible impact on my life, keep it up!

[itsReshad]

Finally! Love your content, you teach and explain really well, I have learned tonnes from following along with you!

[dr.shefalisadvice]

The best thing about you is that you provide all the dataset's that you have used in your videos not like other channels on YouTube who just teach data manipulation but don't give us the data set they have used thank you so much man...

[dasmutterhaus]

Oh Keith we can see you having fun! That is great! Loved the intro

[Kevin-gmgx]

Hee is back !! Thanks Keith - your content is really awesome, thanks for posting this :)

[wahaha]

Great job! This series (solving real world data science problems) really differentiate your video fro m others.

[danieldiaz]

Very informational video, and you make new topics sound less intimidating for beginners. Thanks

[ahmedsamy-ugbv]

thanks,
please keep posting similar videos for real world projects, Its very useful

[himanshuverma]

Can you please do a video on real time ML project involving extensive EDA.

It would be very much helpful and interesting to learn.

[kirubaselvi]

Loved the intro and you are the best in teaching these things✨

[lautaroperez]

Another great video! Greetings from Argentina!

[Paperscissor]

Great vid. Is it possible to use people 's images like actors faces and it tells you their names just by using facial data?

[MM-uhqk]

Great video. Very well done!
Will setting the metric to "F1 score" serve the same purpose as ["precision", "recall"]? Which one works better to prevent false evaluation resulting from class imbalance?

[jacktrainer]

Not using a Mac?! <clutches pearls> Anaconda is definitely the way to go on a Mac. I think they still use it in 6.001x & 6.002x. If there is no conda install command for a package, you can still use pip install w/o a problem. This looks like a fun data set. Looking forward to trying it!

[nosinibnamahbub]

Love your content. Would you please make segmentation related task using pytorch

[dongliang]

Nice work, Keith.
I’m trying to locate this video with one of your playlist channels, but couldn’t find it(especially not in one called ‘data science’). So could you please tell me how I can find the channel for some similar videos?

[luyckr]

Hi Keith, are you using the WSL Manager by Bostrot?

[asmafaraj]

Hallo, I am trying to evaluate submission data set.
1. is it equal to test data set?
2. and how to get the true_labels ?

[mehdismaeili]

Hi, thanks for this great video, you are a very good teacher.