Machine Learning Tutorial Python - 8 Logistic Regression (Multiclass Classification)

[@codebasics]

Check out our premium machine learning course with 2 Industry projects: codebasics.io/courses/machine-learning-for-data-science-beginners-to-advanced

[@Charmingenby]

There are very few teachers who actually make us fall in love with learning. You have an incredibly fascinating way of teaching Sir!!

[@codebasics]

😊👍

[@ritikpratapsingh9128]

I do not usually comment but you wrote the code so simple and explained so beautifully that i had to praise you. Thank you so much !!

[@codebasics]

Complete machine learning tutorial playlist: kzbin.info/www/bejne/nZ7Zp5Sll9Jqm7M

[@anujvyas9493]

I solved the exercise and my model got an accuracy of 96.67% Thanks for making such great videos.

[@sonalgarg5628]

@@anujvyas9493 can you please send the solution..i also got same accuracy but unable to do prediction

[@anujvyas9493]

@@sonalgarg5628 Sure! Email ID ?

[@sonalgarg5628]

@@anujvyas9493 [e-mail]

[@anujvyas9493]

@@sonalgarg5628 Sent it to you! Sorry for the late reply.

[@peiyuankao1213]

Thanks for your teaching! I like your tutorials and exercises, that make me quickly understand.

[@maruthiprasad8184]

I got accuracy 93% for iris data set. Thank you very much to make ML simple.

[@rayithki]

Thank you. I wish I had discovered your channel 6 months ago. I could have saved so much time.

[@PoojaPatel-bi4wr]

Sir , Whatever you teach it's very very interesting and I think I am luckiest person which I am reading from your videos It's very helpful for us and you are great. I have seen many videos but no one teaches like you.

[@Ankurkumar14680]

Great tutorial, thanks a ton for shaing this amazing stuff. Request you to start a series on NLP, Deep Learning or Text Analytics

[@SohamPaul-xy9jw]

Thank You. After watching previous 8 videos, I tried this Iris exercise on my own and my model actually predicted so well, with a score of 1.0

[@satyazigyansu6873]

it is overfitting bro

[@parththakor7362]

@@satyazigyansu6873 accuracy is varies with random state and test size random state = 42 and test size = 0.2 then accuracy = 100% random state = None and test size = 0.3 then accuracy is around 97% and it varies every time

[@yashchaudhari5289]

@@satyazigyansu6873 no brother it depends on dataset whether it is testing or training. If it is on testing dataset then it is not overfitting, if it is is on training dataset then it is overfitting.

[@wsgsantos]

As always great video. Greetings from Brazil!

[@jagjeetagarwal1856]

Within 2 Days I have addicted to this channel......I am on this Channel for around 5-6 hours Continuously....... Please Continue the Series......Thanks

[@jayshreedonga2833]

Thanks sir. Simply you are great for such type of free courses.Nice service to society

[@stanlukash33]

I CAN'T SAY THIS ENOUGH - THANK YOU!

[@pallawkumar9846]

Thank you for these awesome tutorials. Please upload next tutorials.

[@-theworldofscience4153]

The contents are actually very engaging and helps u tolearn complex topics very easily

[@masterofnone754]

probably the best tutorial series for beginner thank you!!!!!

[@codebasics]

Glad it was helpful!

[@nitinsrivastava2136]

Awesome. Thanks for sharing. I love the way you teach topics. So easy to understand. Thanks again.

[@codebasics]

Yup nitin, things don't have to be taught in a hard way.. there is always an easy way to explain the concepts :)

[@tcsanimesh]

Your explanation is at a different level. Just one request please add the different machine learning algorithms a bit fast as once someone starts leading from your channel gets hooked up to it ...

[@nishkarshaggarwal2692]

you are a great teacher.... thank u for this series

[@liangyumin9405]

Nice tutorial, I have forked your project PY .THX

[@rehanabbas4661]

Respect and appreciation from 🇵🇰 . Interesting teaching skill. 👍

[@jyotimalik8960]

finalllllly I understood how to interpret confusion matrix for multiclass classification thankyou!!!!

[@yoniziv]

Very clear, thank you!

[@muskan_salampuria]

One of the best tutorial... Thankyou so much...It is very helpful and informative.... I wish to see more videos on other topics...

[@codebasics]

Glad, you liked it.

[@Kikeina]

A little detail... after updating sklearn to version 0.20.2 or higher it will be needed to specify a solver and multi_class specification as parameters to avoid warning errors. For instance "model = LogisticRegression(solver = "newton-cg", multi_class="auto")"

[@russnagel1]

Thank you very much. You just saved me a big headache. I had the warning and came looking to the comments for help. Great job.

[@Kikeina]

@@russnagel1 Happy to see that the comment is helping somebody. You made my day.

[@dhruvpathak1850]

Very helpful, I tried using max_iter / n_iter to 200, in the model.fit() part, but that didn't work either.. eventually, it's your suggestion that did work!

[@drpebba2679]

my savior

[@dixxanta]

u can also use standard scaler

[@mabelkarani]

at 7:50 , use this >> model = LogisticRegression(solver='lbfgs',class_weight='balanced', max_iter=10000) to avoid this warning >>> 'ConvergenceWarning: lbfgs failed to converge (status=1): STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.'

[@harsh2014]

Thank You Mabel Karani , you helped me !

[@tanvirahmed552]

thanks 😊

[@muhammadtayyab2148]

thks

[@aaditstudent]

@@muhammadtayyab2148 what was ur accuracy score using the above parameters?

[@palurikrishnaveni8344]

Thank you madam

[@kashifahmad93]

Got 96.66% accuracy.....while practicing on your given iris.csv dataset...I am new on your channel, but got addicted to your videos, especially to the playlist of machine learning... please keep teaching us in same way. Thanks a lot..

[@codebasics]

That’s the way to go Kashif, good job working on that exercise

[@RahulRaj-cy1xb]

Bro we need to download exercise from kaggle? As sir only uploaded image on github

[@parththakor7362]

@@codebasics accuracy is varies with random state and test size random state = 42 and test size = 0.2 then accuracy = 100% random state = None and test size = 0.3 then accuracy is around 97% and it varies every time

[@jawadasif8130]

really appreciate your hard work. from your videos it was super easy to learn the concept . thank you

[@codebasics]

You are most welcome

[@pakpoomtivarkornkit8525]

Thank you so much....liked and subscribed.

[@yourlifeonpower]

Another awesome video! Thank you

[@vinodkinoni4863]

thanks for good tutorials

[@rakeshg2926]

Thanks your tutorials are very clear and intutive and easy to understand.

[@codebasics]

Rakesh, thanks for your kind words of appreciation

[@raghuvamsiyaram7248]

I got 96.66 accuracy. Thanks.

[@rishipatel2221]

Thank you so much for this invaluable series

[@codebasics]

Glad you enjoy it!

[@strongsyedaa7378]

@@codebasics Kindly make a video on confusion matrix multiclass classification please 🙏

[@MdAbdurRahaman-f2d]

one thing i don't understand. in the heat map, you said, if the number was not zero, the model accuracy failed there. but in the first example, 37 times I fed my model zero, and my model said it was zero. same as like, 40 times i fed one and my model said it to be one. So, the accuracy is perfect with not being zero in the diagonal part of heatmap. thanks a lot for your marvelous effort

[@rajatpati8808]

Waiting for your next videos. Hope you will upload soon.

[@rambaldotra2221]

Iris dataset -> 97.777777777777 accuracy with test_size =0.3 I have fallen in love with this amazing knowledge 🤩.Thanks a lot Sir ❤️.

[@oilidzaghloul3266]

I got an accuracy of 1 with test_size=0.2.

[@RahulRaj-cy1xb]

Bro we need to download exercise from kaggle? As sir only uploaded image on github

[@nandithass7326]

pls provide the code

[@dennisasamoah2213]

simply amazing

[@mohammedashfaqueaslam2566]

Amazing content you make it all seem easy

[@codebasics]

Glad you liked it Mohammed.

[@vdud007]

Thank you and practice exercises are useful as well

[@codebasics]

Glad you liked the exercises Vishnu

[@nastaran1010]

Thanks a lot, very clear

[@golammuktadir4355]

Good approach for coding the basic machine learning . Carry on

[@codebasics]

👍😊🙏

[@achelias8477]

On my way to watch your whole playlist. You are a great techer! I got accuracy 95.6%

[@codebasics]

👍😊 wish you all the best

[@olutokijohn]

Can you share your solution ?

[@aaditstudent]

What parameters did you use for the LogisticRegression model?

[@safwansalehjee7961]

Is there no Exercise solution?

[@hamzanaeem4838]

Excellent explanation

[@nicojapasmusic]

Very helpful, thanks!

[@codebasics]

Glad it was helpful!

[@javi19841000]

Great job, Thank you ver much

[@codebasics]

Glad you liked it!

[@radhedhabas]

I got accuracy of 96.66%. Thank you so much for your initiative. Best part of your playlist is exercises that give confidence and a clarity how to apply logics in form of code. And best part you talk about practical use cases.

[@parththakor7362]

accuracy is varies with random state and test size random state = 42 and test size = 0.2 then accuracy = 100% random state = None and test size = 0.3 then accuracy is around 97% and it varies every time for best way choose random state = 42 or 10

[@nandithass7326]

pls provide the code

[@radhedhabas]

@@nandithass7326 please look exercise section in given notebook: colab.research.google.com/drive/1ayUBBqEvH-mDkYHVqMFO6-nlWRyZhyNt?usp=sharing, as I'm using sklearn, accuracy varies every time.

[@abhishekkhare6175]

got accuracy of 93.34%. Thanku very much really addicted to your videos.

[@codebasics]

Well done

[@tanoychowdhury6375]

Sir, Please make videos on other topics of machine learning like k nearest neighbour , support vector machines. Your videos are very very helpful. please continue this series.🙏🙏

[@orangewares]

You can refer to videos of sentdex. The videos are much better including k nearest neighbor. kzbin.info/www/bejne/hXjbmKF-bd5lhK8

[@nareshjanjirala472]

nice tutorila.... by watching your tutorials lot of people are opeining institutes in Hyderabad

[@codebasics]

Ha ha.. are you serious? 😊

[@tewatiakuldeep86]

Sir your way of describing things is very easy to grab and understand. Thank you for the tutorial. I request you to please also make a few videos of analyzing data (statistics) before using it into a model. Like variable correlation, and what variable should be used and which one should be dropped, etc.

[@codebasics]

point noted kuldeep and thanks for your appreciation. I want to add lot more content but unfortunately facing health troubles. once i recover I will be back with full force :)

[@nayyershahzad8051]

I got 96.66% accuracy for Iris dataset exercise. Great work! Thoroughly enjoying and learning a lot from your courses.

[@digvijaymahamuni7722]

i got 94.73% does it vary? or have I done any mistakes?

[@fazalahmad1546]

I got 100.0%

[@fazalahmad1546]

@@digvijaymahamuni7722 this is due to a very small dataset

[@mohammedsohilshaikh6831]

@@fazalahmad1546 check for overfitting

[@mdlwlmdd2dwd30]

Hey guys chill it isnt like you guys working in backend developing library. also it is relatively clean dataset already.

[@franky0226]

Thank you so much sir :) I loved the tutorial! , got an accuracy of 97.72 %

[@codebasics]

Great job!

[@nitishkeshri2378]

how?.....I'm getting only 56%

[@bandhammanikanta1664]

Loving your Lectures sir. Could you please use any best deep learning model for this dataset. Or Suggest me one. :)

[@ehsanakbari3519]

that was awesome🤩🤩

[@vishnuvardhan-wq5qi]

my model is 100 percent accurate for iris dataset. thanks for teaching all the topics which are really important in a clean and clear way.

[@harshthummar6360]

I loved this tutorial..! Absolutely awesome...!! i get up to efficiency= 96.6%

[@codebasics]

That’s the way to go Harsh, good job working on that exercise

[@asedaaddai-deseh8152]

Thanks so much for these great tutorials! I wish you would upload the continuation of this playlist faster so we can learn fast.

[@asedaaddai-deseh8152]

@@codebasics Wow, I admire the fact that you're able to make these videos despite your busy schedule. Keep it up!

[@anand.prasad502]

@@asedaaddai-deseh8152 medium.com/trainyourbrain/would-you-read-this-article-or-not-b757d0e26cf8

[@dhruvkantsharma2955]

Amazing lecture! I got an accuracy of 93.3%

[@vishalrai2859]

i got 88% score

[@KushVarma]

getting a score of 1.0, by using newton-cg solver. Default LogisticRegression() shows warning. You can use model = LogisticRegression(solver = 'newton-cg', multi_class='auto') for better training and accuracy.

[@coxixx]

awesome!

[@AnanyaRay-ct8nx]

got 93.33% accuracy. Thank u so much for this playlist..

[@ayushshankarpurkar1417]

i also got 93.33% accuracy can you please tell me how you did it I want to cross check my procedure.

[@ss57hd]

Wow, Your videos are amazing! And i got an accuracy of 96%

[@codebasics]

great. thanks for working on exercise and congrats on getting such a high accuracy score. Good job :)

[@vinays.m6831]

Sir can u send me that code please... I am not getting that so...

[@sujithramanathan3275]

@@vinays.m6831 PFB code. Please let me know if anything is incorrect. import pandas as pd import matplotlib.pyplot as plt from sklearn.datasets import load_iris from sklearn.model_selection import train_test_split from sklearn.linear_model import LogisticRegression iris = load_iris() x_train, x_test, y_train, y_test = train_test_split(iris.data, iris.target, test_size=0.2) irisModel = LogisticRegression() irisModel.fit(x_train, y_train) targetIndex = irisModel.predict(x_test) for i in range(len(targetIndex)): print(iris.target_names[targetIndex[i]]) irisModel.score(x_test,y_test)

[@mansoorahmed4150]

nice explanation. I have one question. What about if we have mix of dependent variables data, like binary as well multiclass v variables, is it fine we apply multiclass regression?

[@pamp3657]

Good video.

[@wasit-shafi]

At 15:33 I thought you are going to say 'plz plz subscribe the channel, like, comment, share... :) Thank you sir for making such a great videos...

[@rajareivan2417]

so it can only take inputs and predict images from the dataset?, how if i want to predict other images that are not from the digit dataset?

[@josiahvanness4037]

Sir, you forgot to upload the iris flower Solution for ML exercise 8 in you github, there is only exercise description at the end of github file, no exercise folder, no solution. Your tutorial is awesome, everybody is following you to practice. Thank you for teaching us and can you please upload the solution, appreciated.

[@RohitSingh-hc8yi]

awesome

[@yoshidasan4780]

Thank you so much sir !I am so so grateful to you for these wonderful tutorials ,hope i can learn even more and faster.Btw i got my accuracy as 97.77 !

[@pranav9339]

Bro I got the same but is it correct? How can accuracy be so high? Please can you explain

[@yoshidasan4780]

@@pranav9339 because the trends are very similar in the test set data too ig and the variance is also low ...that's the reason i think

[@mojojojo1854]

please do tutorials on Computer vision using Tensorflow

[@cindinishimoto9528]

That would be awsome!! ♥️

[@shivadumnawar7741]

I got 100% accuracy for the iris exercise. Sir give more exercise. These are very helpful, thanks a lot sir

[@sejalmittal1326]

didn't you get total no. of iterations reached ??

[@sejalmittal1326]

Can you help me out ??

[@shreyjoshi18]

increase the size of your test data and then check

[@sejalmittal1326]

@@shreyjoshi18 okay .. thanks

[@foreveryour11]

At 12.17 what we predicted was for X_test. Why did we compare the Y_test and X_predictions? Am i understanding it wrong? 😀

[@mlTS7626]

Superb content, liked this very much 12:50, maybe there's a simple mistake that xlabel should be Truth while ylabel should be Predicted, as we have defined cm in that way

[@arjunpukale3310]

How to recognize whether the classification dataset is linear or non linear if there are multiple variables and cannot be plotted?

[@syed1742]

How to visualize decision boundry through plot and how to optimize using log loss function, and whatever you are teaching that teaching everyone.

[@sanooosai]

thank you

[@vijaydas2962]

In both Binary and Multiclass classification example, you have used the SAME algorithm i.e. from sklearn.linear_model import LogisticRegression model = LogisticRegression() Questions- 1. Does it mean, from LogisticRegression perspective, it makes no difference whether it is binary or multiclass classification? 2. Is there any model parameter that we can tweak to define the class boundary. Say, predict the output as - a) "Excellent", if probability> 0.8 b) "Very Good", if probability >0.6 and 0.4 and 0.2 and

[@g00dvibes47]

It’s very late and I’m very tired, but in a nutshell, sometimes you can use regression as a classifier and sometimes binary classifiers are suitable for multiclass classifications!

[@tanmaythaker2905]

Sir done with the assignment. Got 100% train accuracy for iris dataset and also plotted the confusion matrix.

[@perikalasunny5698]

sir in this video i think you took x and y axis reverse in labelling the cause in confusion matrix arguments its x and y respectively right?

[@cindinishimoto9528]

Awesome exercise! I got an accuracy of 97, 77%

[@aditinagar6688]

Can you please provide the solution link as it is not there on github? It would be helpful.

[@cindinishimoto9528]

Hi, @@aditinagar6688​. Please see below: iris = load_iris() print(dir(iris)) df = pd.DataFrame(iris.data, columns=iris.feature_names) print(df.head()) df["target"] = iris.target print(df.head()) df["target"].replace({0: "setosa", 1: "versicolor", 2: "virginica"}, inplace=True) print(df.head(-10)) x = df.drop(["target"], axis=1) y = df["target"] from sklearn.model_selection import train_test_split x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.3) print(len(x_train)) print(len(x_test)) from sklearn.linear_model import LogisticRegression model = LogisticRegression() model.fit(x_train, y_train) print(model.score(x_test, y_test)) print(model.predict(x_test)) #print(y_test) print(model.predict([[4.9, 3.0, 1.4, 0.2]])) #setosa y_predicted = model.predict(x_test) from sklearn.metrics import confusion_matrix cm = confusion_matrix(y_test, y_predicted) print(cm) import seaborn as sn plt.figure(figsize=(10, 7)) sn.heatmap(cm, annot=True) plt.xlabel("Predicted") plt.ylabel("Truth") plt.show()

[@tejobhiru1092]

@@cindinishimoto9528 thank you so much..! this helps a lot.. i was not able to figure out how to handle that dataset!

[@cindinishimoto9528]

@@tejobhiru1092 ^_^

[@haiderkhwaja2514]

@@cindinishimoto9528 i also need this exercise code very badly

[@hameedali2963]

great

[@nilupulperera]

Dear Sir Very interesting exercise. Model accuracy varies from 0.8 to 1.0, each and every time after a fresh run of the full code (as you explained). The average accuracy is around 9.66667. Thank you very much

[@pranavkushare5109]

use (shuffle=False) in train_test_split()

[@parasf09]

accuracy=93.3% thankyou sir

[@87prak]

You should have talked about what scikit logistic regression is doing under the hood for multiclass. The tutorial does not touch on that and simply runs the program as if it were binary class problem. Is it using n binary classifiers or a softmax, that is what should have been discussed here.

[@muhammedrajab2301]

sir I have done the Iris flower exercise according to what I have learnt from you. I got an accuracy of 1.0 (I thing it is 100%) ! I just done everything according to what I have learnt from you!

[@codebasics]

Perfect and keep it up. The dataset is small hence getting accuracy of 1 is not unusual

[@vedanthbaliga7686]

if you have given random_state or shuffle=True then the accuracy will be 1

[@oilidzaghloul3266]

@@vedanthbaliga7686 even without a random_state or shuffle it s still possible to get 1, it s all due to the fact that our dataset here is small

[@leooel4650]

Can't express how grateful i am to you Sir. I am very willing to even pay for your stuff and help you somehow. Thanks once again, my accuracy was about 92%

[@zerostudy7508]

how much ratio test/train u use? i got 91...%

[@leooel4650]

@@zerostudy7508 20% of the data to be tested. But the accuracy depends as we are getting random data to be trained or tested. My opinion is that your model is correct, we just have different trained data.

[@zerostudy7508]

@@leooel4650 Thank you so much buddy, i just checked that if i use 90% data for train and 10% data test i get 88-90% acuracy, but when i use 80% the data for training i got everage more than 90-100% accuracy. i'll tell you when i figured something out....

[@leooel4650]

@@zerostudy7508 happy to help as I am still figuring things out.

[@zerostudy7508]

@@leooel4650 i got it it something to do with sample and population' if test A=20 and test A=10 then they both got just 1 wrong answer A and B Standard Deviation Sample are A=0.217944947 B=0.316227766 about 10% difference in a nutshell its sound like this: your teacher give 10 questions for exam and your friend got 100, if both of you had 1 wrong answered in the exam, which of you have the highest test score ? have a nice day

[@swL1941]

100% accuracy. Thank You Sir.

[@cbbakshi2020]

Amazing tutorial:) How to make roc_curve for this multiclass problem?

[@AlonAvramson]

Thank you for explaining this in such a nice and easy way. BTW, I downloaded the whole GIT files but could not find the exercise solution for this session, so If some one has a clue please let me know.

[@aravinthm8328]

Yaa the answer for this exercise is not in the file. I solved the exercise, you can also try in the same way as in the model problem. but in the Handwritten digit problem, i got an error when fitting the model :( , i cant correct the error. It showing 'str' object has no attribute 'decode'. Can you help me to come out from this.

[@RahulRaj-cy1xb]

Bro we need to download exercise from kaggle? As sir only uploaded image on github

[@kalaipradeep2753]

I got 100% accuracy🎉🤩

[@asitkmr487]

Very informative. As per my understanding LR model predicts the binary classification problem. It would be great if you can share how this predicts this multi class problem?

[@codebasics]

Check machine learning tutorial playlist on my channel. I have example for binary classification as well and in fact this particular tutorial is for multiclass classification

[@thannasip8001]

Hi ,as you said sigmoid function will convert values to 0 or 1 ,how is it possible to predict digits with this concept ?,for binary output I got it but for digits it confused me

[@ansh6848]

I have set max_iter to 1,00,000 and achieved acurracy of 100%

[@anandsingh1011]

Your all video on any topic have deep theoretical explained with notebook , Can you suggest good resource or book for Machine Learning ?

[@orangewares]

kzbin.info/www/bejne/hXjbmKF-bd5lhK8

[@naveenkalhan95]

@12:47 maybe not that important.. but just for my clarification, I would like to confirm... should plt.xlabel not be 'Truth' and plt.ylabel be 'Predicted' ? Thank you for your hard work.

[@himanshugarg7113]

Yes, it is. It has produced merely transpose.

[@snehagupta-xz1fs]

even I've the same doubt

[@jaiprathapgv2273]

How can use this model used to recognize a new target image out of digits library?how to view the classification graph?

[@musthakhahammed6535]

I got 100% accuracy with 20% test size 😍😍😍 A big salute to this teacher