Real-World Python Neural Nets Tutorial (Image Classification w/ CNN) | Tensorflow & Keras

[@KeithGalli]

Hey all! Been a little while since I have posted, but trying to get back into the swing of things. Hope everyone is doing well. Based on a lot of feedback I have gotten, my current plan is to make a web scraping tutorial with beautiful soup next. Not quite sure what will come after that. Does anyone have any suggestions? Would love to hear them :).

[@thebeston6710]

sup keith galli, i need help trying to fix a bug in this code. This is a game where there is a player and it has to dodge the enemies. But for some reason I still die if i haven't touced any of the enemies. How do you fix that bug? import turtle import random import time from random import randint wn = turtle.Screen() wn.bgcolor('black') wn.setup(600, 800) x = random.randint(-290, 290) y = random.randint(395, 500) x1 = -290 x2 = 290 y1 = 395 y2 = 1000 pm = 10 em = 5 ed = 44 s = 1 score = 0 p = turtle.Turtle() p.hideturtle() p.penup() p.speed(0) p.color('red') p.shape('circle') p.shapesize(stretch_wid = 2, stretch_len = 2) p.goto(0, -350) p.showturtle() enemy_colors = ['darkblue', 'cyan', 'green', 'yellow', 'purple', 'pink', 'brown', 'white', 'orange', 'lightgreen', 'blue'] random_color = random.choice(enemy_colors) e1 = turtle.Turtle() e1.hideturtle() e1.color(random_color) e1.penup() e1.speed(0) e1.shape('square') e1.shapesize(stretch_wid=2, stretch_len=2) e1.goto(x, y) e1.showturtle() e1.right(90) random_color = random.choice(enemy_colors) x = random.randint(x1, x2) y = random.randint(y1, y2) e2 = turtle.Turtle() e2.hideturtle() e2.color(random_color) e2.penup() e2.speed(0) e2.shape('square') e2.shapesize(stretch_wid=2, stretch_len=2) e2.goto(x, y) e2.showturtle() e2.right(90) random_color = random.choice(enemy_colors) x = random.randint(x1, x2) y = random.randint(y1, y2) e3 = turtle.Turtle() e3.hideturtle() e3.penup() e3.speed(0) e3.color(random_color) e3.shape('square') e3.shapesize(stretch_wid=2, stretch_len=2) e3.goto(x, y) e3.showturtle() e3.right(90) random_color = random.choice(enemy_colors) x = random.randint(x1, x2) y = random.randint(y1, y2) e4 = turtle.Turtle() e4.hideturtle() e4.penup() e4.speed(0) e4.color(random_color) e4.shape('square') e4.shapesize(stretch_wid=2, stretch_len=2) e4.goto(x, y) e1.showturtle() e4.right(90) random_color = random.choice(enemy_colors) x = random.randint(x1, x2) y = random.randint(y1, y2) e5 = turtle.Turtle() e5.hideturtle() e5.penup() e5.speed(0) e5.color(random_color) e5.shape('square') e5.shapesize(stretch_wid=2, stretch_len=2) e5.goto(x, y) e5.showturtle() e5.right(90) random_color = random.choice(enemy_colors) x = random.randint(x1, x2) y = random.randint(y1, y2) e6 = turtle.Turtle() e6.hideturtle() e6.penup() e6.speed(0) e6.color(random_color) e6.shape('square') e6.shapesize(stretch_wid=2, stretch_len=2) e6.goto(x, y) e6.showturtle() e6.right(90) random_color = random.choice(enemy_colors) x = random.randint(x1, x2) y = random.randint(y1, y2) e7 = turtle.Turtle() e7.hideturtle() e7.penup() e7.speed(0) e7.color(random_color) e7.shape('square') e7.shapesize(stretch_wid=2, stretch_len=2) e7.goto(x, y) e7.showturtle() e7.right(90) random_color = random.choice(enemy_colors) x = random.randint(x1, x2) y = random.randint(y1, y2) e8 = turtle.Turtle() e8.hideturtle() e8.penup() e8.speed(0) e8.color(random_color) e8.shape('square') e8.shapesize(stretch_wid=2, stretch_len=2) e8.goto(x, y) e8.showturtle() e8.right(90) random_color = random.choice(enemy_colors) x = random.randint(x1, x2) y = random.randint(y1, y2) e9 = turtle.Turtle() e9.hideturtle() e9.penup() e9.speed(0) e9.color(random_color) e9.shape('square') e9.shapesize(stretch_wid=2, stretch_len=2) e9.goto(x, y) e9.showturtle() e9.right(90) random_color = random.choice(enemy_colors) x = random.randint(x1, x2) y = random.randint(y1, y2) e10 = turtle.Turtle() e10.hideturtle() e10.penup() e10.speed(0) e10.color(random_color) e10.shape('square') e10.shapesize(stretch_wid=2, stretch_len=2) e10.goto(x, y) e10.showturtle() e10.right(90) pen = turtle.Turtle() pen.hideturtle() pen.penup() pen.goto(0, 360) pen.color('white') def move_right(): x = p.xcor() x += pm if x > 290: x = 290 p.setx(x) def move_left(): x = p.xcor() x -= pm if x < -290: x = -290 p.setx(x) def move_up(): y = p.ycor() y += pm if y > 390: y = 390 p.sety(y) def move_down(): y = p.ycor() y -= pm if y < -390: y = -390 p.sety(y) wn.listen() wn.onkeypress(move_left, 'Left') wn.onkeypress(move_right, 'Right') wn.onkeypress(move_up, 'Up') wn.onkeypress(move_down, 'Down') run = True while run: wn.update() em += 0.04 e1.fd(em) e2.fd(em) e3.fd(em) e4.fd(em) e5.fd(em) e6.fd(em) e7.fd(em) e8.fd(em) e9.fd(em) e10.fd(em) if e1.ycor() < -395: x = random.randint(x1, x2) y = random.randint(y1, y2) random_color = random.choice(enemy_colors) e1.color(random_color) e1.goto(100000000000, 100000000000) e1.goto(x, y) score += 1 pen.clear() pen.write('{}'.format(score), align='center', font=('courier', 24, 'normal')) if e2.ycor() < -395: x = random.randint(x1, x2) y = random.randint(y1, y2) random_color = random.choice(enemy_colors) e2.color(random_color) e2.goto(100000000000, 100000000000) e2.goto(x, y) score += 1 pen.clear() pen.write('{}'.format(score), align='center', font=('courier', 24, 'normal')) if e3.ycor() < -395: x = random.randint(x1, x2) y = random.randint(y1, y2) random_color = random.choice(enemy_colors) e3.color(random_color) e3.goto(100000000000, 100000000000) e3.goto(x, y) score += 1 pen.clear() pen.write('{}'.format(score), align='center', font=('courier', 24, 'normal')) if e4.ycor() < -395: x = random.randint(x1, x2) y = random.randint(y1, y2) random_color = random.choice(enemy_colors) e4.color(random_color) e4.goto(100000000000, 100000000000) e4.goto(x, y) score += 1 pen.clear() pen.write('{}'.format(score), align='center', font=('courier', 24, 'normal')) if e5.ycor() < -395: x = random.randint(x1, x2) y = random.randint(y1, y2) random_color = random.choice(enemy_colors) e5.color(random_color) e5.goto(100000000000, 100000000000) e5.goto(x, y) score += 1 pen.clear() pen.write('{}'.format(score), align='center', font=('courier', 24, 'normal')) if e6.ycor() < -395: x = random.randint(x1, x2) y = random.randint(y1, y2) random_color = random.choice(enemy_colors) e6.color(random_color) e6.goto(100000000000, 100000000000) e6.goto(x, y) score += 1 pen.clear() pen.write('{}'.format(score), align='center', font=('courier', 24, 'normal')) if e7.ycor() < -395: x = random.randint(x1, x2) y = random.randint(y1, y2) random_color = random.choice(enemy_colors) e7.color(random_color) e7.goto(100000000000, 100000000000) e7.goto(x, y) score += 1 pen.clear() pen.write('{}'.format(score), align='center', font=('courier', 24, 'normal')) if e8.ycor() < -395: x = random.randint(x1, x2) y = random.randint(y1, y2) random_color = random.choice(enemy_colors) e8.color(random_color) e8.goto(100000000000, 100000000000) e8.goto(x, y) score += 1 pen.clear() pen.write('{}'.format(score), align='center', font=('courier', 24, 'normal')) if e9.ycor() < -395: x = random.randint(x1, x2) y = random.randint(y1, y2) random_color = random.choice(enemy_colors) e9.color(random_color) e9.goto(100000000000, 100000000000) e9.goto(x, y) score += 1 pen.clear() pen.write('{}'.format(score), align='center', font=('courier', 24, 'normal')) if e10.ycor() < -395: x = random.randint(x1, x2) y = random.randint(y1, y2) random_color = random.choice(enemy_colors) e10.color(random_color) e10.goto(1000000000000, 100000000000) e10.goto(x, y) score += 1 pen.clear() pen.write('{}'.format(score), align='center', font=('courier', 24, 'normal')) if p.distance(e1) < ed: time.sleep(s) print('Score:', score) quit() if p.distance(e2) < ed: time.sleep(s) print('Score:', score) quit() if p.distance(e3) < ed: time.sleep(s) print('Score:', score) quit() if p.distance(e4) < ed: time.sleep(s) print('Score:', score) quit() if p.distance(e5) < ed: time.sleep(s) print('Score:', score) quit() if p.distance(e6) < ed: time.sleep(s) print('Score:', score) quit() if p.distance(e7) < ed: time.sleep(s) print('Score:', score) quit() if p.distance(e8) < ed: time.sleep(s) print('Score:', score) quit() if p.distance(e9) < ed: time.sleep(s) print('Score:', score) quit() if p.distance(e10) < ed: time.sleep(s) print('Score:', score) quit() wn.mainloop()

[@SSSNIPD]

I have done andrew ng's machine learning course. It would be awesome if you could explain how to apply those concepts to python and start machine learning in python.

[@arhataria]

Another series of real "business" world data science tasks (using pandas and machine learning together!) would be awesome :D I'm from a non-IT background and the last series you posted actually helped me a lot at my work!!! Thanks a lot :)

[@thebeston6710]

Can you try to do OpenGL

[@CassieYT1]

Maybe a video on GANs would be cool?

[@pkPatricia]

Keith cannot thank you enough, there are places that present this topic in like 6 hours or something and you got to the point and really filled in some gaps I had with this concept. Very nice!

[@afrozahmad3173]

Keith u finally posted u dont know how much i wait for your tutorials man and thx for the NN vid very good and u explain very well.Stay safe at home guys

[@KeithGalli]

Sorry for keeping you waiting! I appreciate your support 🙌

[@omkarsatapathy8209]

Great job body. You made me love CNN 😃. I am accepting you as my teacher ☺️🙏

[@satchsumner-waldman8761]

Keith, your videos are unbelievably good, I've watched almost one per day for a week straight. Thanks so much for putting out this content!

[@CodeEmporium]

Gotta love those details. Awesome work!

[@KeithGalli]

Thanks man! Big fan of your videos :)

[@jasminehuang3611]

Hi Keith, really, can't thank you enough for making all the amazing videos that's perfect and easy to understand for someone like me who's first language isn't English :)

[@KeithGalli]

Glad that you enjoy them and happy to hear my English is easy enough to understand! :)

[@sireechan]

Thank you soo much, your explanation is relatable and easy to understand! :) I can't believe you are teaching for free. Happy to learn from you ☺️

[@renipratiwi5720]

i gladi find your channel. now i know how to resolve my main problem in my rock-paper-scissors exam. youve got a new subs!

[@seekergx8096]

Thank you so much man! I've been studying ML like crazy and even so it's hard to get going without seeing someone do it first, you're really helping me out!

[@ajayrana4296]

this is how teaching should be... love u brother

[@tamoorkhan3262]

The most illustrative tutorial on TensorFlow & Keras regarding NN and CNN. (Y)

[@cagataykirici9878]

It has now become an awesome rock-paper-scissors series man, thank you!

[@saadbenbrahim954]

Thank you! My man for this tutorial. A watcher from Morocco

[@KeithGalli]

You're welcome :). Would love to visit Morocco some day!

[@saadbenbrahim954]

@@KeithGalli you're welcome, I will be happy to show you around.

[@wt7658]

It is very helpful for my preparation for the Tensorflow cert exam. Thanks very much!

[@wiz8058]

Finally Keith brought us our favorite food. Thanks 🙏🏼. Love how you made it too. Best tuition and keep it up bro

[@richarddankwah3088]

Hey !!! Mr. Galli is been a while, Infact your Tutorials on Data Analysis with python on that Mobile electronics and appliance was superb. It helped me to complete my project. Thou I found out few lapses and was a challenge which I was able to solve them myself and I did like a charm. You gave me that boldness to apply for few Jobs with my new project. Thank you. But the lapses really strengthen me very well and made me exploit more on different ways of writing codes. "But there is One last step you failed to do with that data, that is Predicting future sales with the data. Please kindly come back to fill that gap for us please . "Using Tensorflow and Keras with LSTM and or ARIMA model to predict future sale on items and dates". Still in waiting, you already have 3 stars with me, waiting for the part 2, (Using Tensorflow and Keras with LSTM and or ARIMA model to predict future sale on items and dates) which will be 2 stars to complete a 5 Star. Stillll In waiting !!!!!!!!!!!!!!!!!!!!!!

[@Mei-ChiShih]

May I have a question？ Why my terminal shows " cannot reshape array of size 226800000 into shape (372,300,300,1)" while I ran this code"train_images = train_images.reshape(2520, 300, 300, 1) test_images = test_images.reshape(372, 300, 300, 1)".

[@MrKrishnalovesyou]

Stay safe my dear friend Keith, I m Indian & here covid cases are increasing 🇮🇳

[@KeithGalli]

Sorry to hear that covid cases are increasing over there :/. Hope you stay safe & healthy as well. We all will get through this!

[@MrKrishnalovesyou]

@@KeithGalli Yes Keith... We will fight & win ✌️

[@prashantbisht2219]

@@MrKrishnalovesyou Bht hi ache bhai, mjaa hi aagya

[@lucky4the4turtle]

You're Back!!!!! Dude we've missed you!!

[@quytrannhat11]

extremely good, easy to understand, good job bro

[@qingdekueh7186]

Hi, how do you use this to predict a new image if it is rock paper or scissor? :(

[@RamiroAsincrono]

Thank you! From Argentina

[@irumidesu9236]

dude, thanks for explaining most of the basics!

[@GeorgePap99]

Bro, you help a lot with your vids!

[@Mei-ChiShih]

Excellent video

[@realneosi]

HOW DID I MISS THIS UPLOAD 🤩

[@vincenrow7190]

You made it so clear step by step, great tutorial, best so far. Just wonder if u could make a video about custom object detecting using tensorflow keras on google colab, I've searched all around the youtube, all of the videos about this topic is kinda frustrating and dispointing, they teach people in a complex way, i'm a lazy man, really like the way you teaching 🍺

[@Kingsohio]

This is dope bruh! Thanks for video

[@Veguinho]

Great tutorial! Thanks dude!

[@phan5001]

Thank you so much! This is very helpful for my project!

[@lokguanlim7420]

Learned a lot. Thank you so much!

[@KeithGalli]

You're welcome!

[@lorryzou9367]

The pooling size of the MaxPooling layer is 2, which is too small to represent the image. If you set the pool_size to (5, 5), the test accuracy will be 80%.

[@manu93ize]

Congrats man 🙏 finally got monetized .

[@ВячеславЗахаров-х1ы]

Thank you for the great video!!!

[@moseskioko7820]

good stuff, thanks for the sharing knowledge

[@prakharkaushik5200]

So how do I use this model to make predictions

[@hawaabasheer5239]

Where can I find the series of Keras and tensorflow videos?

[@haseenabadulla3208]

Can you please explain how to load the data if saved locally but in different folders, that is, test images in one folder, train images in folders 0, 1 and 2 as per their types? Thank you.

[@hungnguyenngocduy2088]

Hi Keith, If I want to use a dataset that is already on my computer instead of downloaded from the Internet, how do I do it! Much thanks :-)

[@JO-ej3nh]

I was trying to use the cat and dog dataset but the shape on shows the size of the samples and nothing else. Also there is only train data and only test data. Any solutions?

[@digigoliath]

Awesome!! TQVM!

[@notrevealable]

Thank you so much!!

[@dhainik.suthar]

How to use this model for real time image classification ?

[@andrehoffman6887]

Always good to hear you, Keith. By the way, what'd you recommend for people to get a first job using CNN's? Thanks a lot

[@woonie3134]

How can u extract the label of a classified image when working with a large dataset? I would like to classify food image and recommend recipes based on the label identified but I am not understanding how to do this. I WOULD kindly ask a tutorial on this matter. There is barely information on food related projects plz

[@YazeedAlkosai]

fruitful video, hope to see more in the 3D shape model using CNN with Keras. what if I have an object of 3D and need to train and predict the feature shape as measuring the shape by using CNN prediction and accuracy ?

[@jongcheulkim7284]

Thank you so much^^

[@davidadu1113]

Your Pandas tuto was so great. It has urged me to go further. Can you do a tuto on how to deploy a machine learning model in django?

[@joseph3036]

That is easily done using Streamlit, checkout this video. m.kzbin.info/www/bejne/l3K0lmumoaysZ68

[@parpithovas]

thank you. great job!

[@aisham.yousuf5772]

how to load our own dataset (pictures/images of objects) instead of using one of the existing dataset of tensorflow?

[@KeithGalli]

I personally like using an ImageGenerator & the flow from directory methods in Keras. Here is a link you can read more about that: datascience.stackexchange.com/questions/65979/what-is-the-correct-way-to-call-keras-flow-from-directory-method

[@WhiteError37]

Keith, is it possible to predict where something is via a signal strength value and it’s lat and lng? So the higher the signal strength the closer you are to it. I’m struggling to write this in code form basically like a weighted median value but the Ai does the work and adjusts the weights. All that the model needs to output is a lat and lng of where it thinks the signal is coming from.

[@thezeroquotient]

Insightful!

[@mohamadalemi2596]

is there any paper base on your tutorial? I want use it for my university project but it should be relative to specific paper (nice to be in ISA)

[@dab30012151227]

Thank you Keith for such a lovely video. While running the code "model.evaluate(test_image, test_label)", I got below error. I am not sure why train and test size should be same. "ValueError: Data cardinality is ambiguous: x sizes: 2520 y sizes: 372 Please provide data which shares the same first dimension". I checked the shape of train_images and test_images and they are same "(300, 300, 1)"

[@Nadelaev]

Thanks for your videos, man! But can you do some of PyTorch examples?

[@KeithGalli]

That is something that I want to do, but I don't have it on the immediate roadmap. Think it would require a decent amount of planning to do properly and want to focus on some other video topics before diving into PyTorch.

[@Nadelaev]

@@KeithGalli gonna keep waiting! appreciate your work)

[@ayencoscolfield3312]

Same here too eagerly waiting for a deep learning pytorch video examples from our man Keith Gaili

[@faizrazadec]

While showing the example using "fig = tfds.show_examples(info, ds_train)" It'll not work, it'll just throw an error, this signature has been modified from (info, ds_train) to (ds_train, info), so change it in your code...

[@jackoverstreet2835]

If anyone is using Anaconda and importing the TensorFlow Dataset through pip isn't working then try: conda install -c anaconda tensorflow-datasets in the Anaconda Prompt, it worked for me.

[@brandonruiz1746]

hey keith hope you are well bro. I was hoping you could do a video on this same sort of stuff but doing it by using google drive. I have datasets for training images and testing images in my drive with a csv for each containing their forresponding labels. Thoughts?

[@KeithGalli]

I have a similar video that is already filmed going over similar stuff. It got delayed by the sponsor so I'm not a exactly sure when I can post it

[@DanielWeikert]

Thanks. Why exactly did you keep the red color channel?

[@KeithGalli]

Ultimately we could have kept any of the color channels and the performance would remain roughly the same. Each of the color channels would produce a pretty similar grey scale image. No real reason I took red specifically other than I thought it made sense to just use index 0.

[@DanielWeikert]

@@KeithGalli ok thanks, so far I have always seen converting the color image to rgb upfront so I was wondering.

[@seraf_in_._._._._._._._._._._]

Thank you for the nicely explained video! Helped me a lot. One question: If i use 64 filters in the first layer and 32 in the second (I'm talking about the net architecture around min 30), will i have 64*32=2048 filters in the second layer? Thanks!

[@igordovgan7512]

Thank you

[@tinsp253]

Hi Keith, great job. I have learned a lot from your videos. Hope you will produce more. By the way, I have followed your video and the image classification work as expected. How can I predict one single image? When I pass in test_images[0], the predict function return "... expected ndim=4, found ndim=3 ...." obviously it is expecting 4 dimensional array.

[@KeithGalli]

I think you could pass it in as predict(np.array(test_images[0])) and it will work. Glad you have liked some of my videos! :)

[@tinsp253]

@@KeithGalli I do this: print("Predict: ", np.round(best_model.predict(np.array(test_images[0])))) and the result is still the same as below "ValueError: Input 0 of layer sequential is incompatible with the layer: expected ndim=4, found ndim=3. Full shape received: [None, 300, 1]"

[@azzmaniac4139]

You should try this : best_model.predict(np.array(test_images))[your_number] and it'll work!!

[@saketshetty]

@@azzmaniac4139 When I'm prediction the model based on how you told it's returning an ndarray instead of the label, any idea how to tackle it?

[@sdiabr6792]

I can't thank you enough man Really amazing stuff Quick question Do you have any competitive programming background? And if so, could you give some tips on how to get better at it? Thanks so much man

[@aresagathos2075]

Mnist example link in description points to 404 page

[@siddharthmani9588]

Why are you using epochs = 5? aren't they a little less?

[@joseph3036]

To save some time.

[@siddharthmani9588]

@@joseph3036 Is there an ideal number for the total epochs one must run?

[@joseph3036]

10 epochs have worked well for me so far, too many epochs will cause overfitting. stackoverflow.com/questions/47421951/keras-optimal-epoch-selection

[@soybeanz8582]

keith please make a video about the drawing app i send its cool it took hours to make

[@soybeanz8582]

do the drawing app i put the code

[@soybeanz8582]

keith i send you a code on your turtle graphics video its a drawing app can you put it in a video