model = SVC(kernel = 'rbf', C = 4, gamma = 'scale') With the above config, I got a model score of about 99.17%. Test size was 20%, as mentioned. Thank you, these tutorials are amazing! :) cheers!
Thank you very much for these videos. They are really helpful. I did the exercise and got 99% when C=4. Any increase in C did not affect the accuracy. Also, any alteration made to gamma and kernel dropped the accuracy drastically. Thank you once again.
Your teaching skills are unmeasurable and it's very easy to understand no need to scratch our head for looking at some other training institute. I have executed load_digits datasets and found the following score: For 'rbf' kernal, score -98 'linear' kernal, score -97
Thank you for this. They are really helpful. I did the exercise and got 99.17% when C=10. Any increase in C did not affect the accuracy. Also, any alteration made to gamma and kernel dropped the accuracy drastically. Be blessed.
Excellent video, I'm doing a review of what i learned a year ago in a deep learning course in the university (i'm a geophysics graduate) with this playlist without seeing too much math. For C = 25 kernel = rbf and gamma = scale, Test_size = 0.2 Accuracy = 99.70%
I used model = SVC(C=2.0, gamma='auto', kernel='rbf') and got an accuracy of 100% Can you check that it is right or not? Also I used random_state = 100 in train_test_split method for random values
Hello sir, thank you for your videos. It really helps from the beginner of the video which you have listed in data science playlist. 😄 The model in default method is 99.65% in train and 99.4% in test. Whereas gamma method will lower down the accuracy of the model from 99.4% to 75% therefore it has explicit shows the gamma method is unsuitable for the scenario however the regularisation has improve the train set to 1 and testing set is retained the best accuracy of model. Yet, kernel parameter as linear has also provided a good accuracy of model. Thank you for your guidance.
Tried couple of iterations finally I got 99.166% accuracy with all default parameters. random_state=1 while defining train test data.. Thanks a lot Sir
I got score 99.4 when c=1 and gamma=scale And i got 50 when gamma = auto And 99.7 when gamma = auto and c=10 Thank you sir for this series. And following the tutorial with doing exercises
In linear kernel score is 96.9 percent and in rbf kernel score is 40 percent... With gamma value the score is 0.06... And with the regularization value the score is around 45. 83 percent
@codebasics in the exercise solution of digits datasets, you have created a dataframe from data & target, then added target, then removed target. Does it bring any added value? What I did was just digits = load_digits(), train_test_split(digits.data, digits.target, train_size=0.8) , and do a SVM training on X train and y train.. still giving an accuracy of 98.05 (linear) 98.33 (rbf).
I intentionally did that just to show features and target variable side by side in a same table. Other than that you are right that it doesn't add any value.
@@codebasics aah okay. thanks. i tried starting machine learning many times, but never found as simple tutorials as yours. thanks a lot for the videos.
Great! Sir, Can you elaborate something about plotting the hyperplane (the decision function) in matplotlib I want to see the best line which classifies the data
1)What is good to have- a large gamma or regularization parameter? 2)We used only fit() but not fit_transform(), is it because the rbf Kernel will perform the transformation itself to scale the features and the target labels?
Thanks for your effort sir, but there is something I wonder.. When I fit a model, I can't see any description like you have in your jupyter notebook.(C=1, cache_size=200 etc..) I can't see them.. is there any way to see them?
used hyperparameter tuning here to get 100% for train and 99.72% for test...luckily data was clean cause im not very experienced in data cleaning and here i didnt even do too much data visualization
Why we are not using One hot encoding technique to transform categorical variables into binary variables in classification videos? I could have used logistic regression right?
Hello sir, thank you so much for this video. I got 99.25% when i put C=1. If, I use kernel='rbf' then, I got 99 % accuracy. And for kernel='linear', I got 97.7% accuracy.
Usually we predict the values based on x_test value and we get some predictions for x_test values. And after that we compare the y_pred with the y_test, but in the video the predict code is not written or is it like that only ??
As in this, you have given the last example for practice that same example we have solved in the Logistic regression model!! then what will be the difference b/w them?? I am talking about (Load Digits problem)
got a score of 99.16 for my test samples with C=any thing more than 2( i wonder why there were no difference between C=2 and C= 100, i got the 99.16 accuracy for all the values for C more than 2!). didn't change the gamma or the score would be destroyed! the kernel ='rbf'. thanks for this amazing tutorial BTW! :)
why we shouldn't use decision tree for this iris data set since it can also classify the points well as you shown in scatter plot. pls clarify my doubt?
well you can use decision tree as well. It was just that this tutorial was for SVM hence I used svm here. I didnt say svm is the best method to solve iris classification problem
hello great videos, loved this series. Can you please do a video on imbalanced data sets in classifications problems? Maybe just add onto a previous example you have but with a case where there are very few "1" or "true" values compared to "0" or "false" . thanks for you consideration!
What if we do sliding window to get dimensionality reduction (reduce the length of data), then how to classify the data if the length of data and target is different?
Thank you sir for this tutorial. Sir in the In[21] it is showing the error like cannot import the name 'trapezoid' from 'sklearn.utils.fixes'. Maybe it is due to the new version. What should I do?
**HELP** In my jupyter notebook, i don't get the information of modules like svc(*, c=1.0, kernel='rbf', degree=3, gamma='scale', coef0=0.0, shrinking=true, probability=false, tol=0.001, cache_size=200, class_weight=none, verbose=false, max_iter=- 1, decision_function_shape='ovr', break_ties=false, random_state=none) whenever I execute the model with X and y train datasets. What I get is SVC(), everything blank inside. so what should I do to get all the information????
When I did the exercise, rbf performs slightly better for me than linear. I believe when you created your notebook, the default gamma was auto. Using the scale option provides much better results than auto for rbf.
Hi. I have a aquestion, when we execute the model iteratively, the score of model is chenging. So, How can we say , what is out accuracy of model, for example. in papers as a result?, please give me some clues, 🙏🙏🙏🙏🙏🙏🙏🙏🙏🙏
I think it should be high C corresponds to low regularization, which means the classifier don't penalize too much on classification error. Vice versa for low C.
Check out our premium machine learning course with 2 Industry projects: codebasics.io/courses/machine-learning-for-data-science-beginners-to-advanced
model = SVC(kernel = 'rbf', C = 4, gamma = 'scale')
With the above config, I got a model score of about 99.17%. Test size was 20%, as mentioned.
Thank you, these tutorials are amazing! :) cheers!
again great job sagnik. I am seeing that you are on the roll and finishing all the exercises from this playlist. keep it up :)
@@codebasics high model score leads to overfitting? as I got 98% model score with 60% training size
I got 99.25% model score with 70% training size
my model is overfitting dude, 1.0 accuracy.
Got 1.0 score when C=4 for iris data set. Thank you Sir! Your machine learning Playlist is a boon for beginners like me.
That's not always a good thing though. In most real life problems, that would mean that your model has become overfitted.
Same here😱
@@nikitakazankov4099 Ikrt😏
@@nikitakazankov4099 Though it does make sense, Whenever I see a Russian name I bow down because of their intelligence.
@@nikitakazankov4099 bro, The accuracy is on test dataset. if it's on training dataset then it must be overfit
Thank you very much for these videos. They are really helpful. I did the exercise and got 99% when C=4. Any increase in C did not affect the accuracy. Also, any alteration made to gamma and kernel dropped the accuracy drastically. Thank you once again.
Thank you so much for your presentation. I have learned a lot.
Exercise
Test size=0.2, C=1, kernel='poly
Accuracy: 99.17%
Your teaching skills are unmeasurable and it's very easy to understand no need to scratch our head for looking at some other training institute.
I have executed load_digits datasets and found the following score:
For 'rbf' kernal, score -98
'linear' kernal, score -97
Siddu, thanks for complement and good job on exercise. 👏👏👏 That is indeed a nice score
Thank you for this. They are really helpful. I did the exercise and got 99.17% when C=10. Any increase in C did not affect the accuracy. Also, any alteration made to gamma and kernel dropped the accuracy drastically. Be blessed.
one of the best lecture I have ever watched
Hey Abhishek.
Great thanks for your kind words.Stay in touch for more videos and share our channel if you really find it worth.
This series is the best I have seen on simple and explicit Machine learning and Algorithm.Thanks
Glad to hear that!
Wow! how brilliantly working and good teaching method as well . thx sir from Pakistan ... keep it Up!
**** pk
Excellent video, I'm doing a review of what i learned a year ago in a deep learning course in the university (i'm a geophysics graduate) with this playlist without seeing too much math.
For C = 25 kernel = rbf and gamma = scale, Test_size = 0.2
Accuracy = 99.70%
I used kernel = linear and it gave me an accuracy score of 1.0 :3
Exercise solution: github.com/codebasics/py/blob/master/ML/10_svm/Exercise/10_svm_exercise_digits.ipynb
Complete machine learning tutorial playlist: th-cam.com/video/gmvvaobm7eQ/w-d-xo.html
I used model = SVC(C=2.0, gamma='auto', kernel='rbf') and got an accuracy of 100%
Can you check that it is right or not?
Also I used random_state = 100 in train_test_split method for random values
Hello sir, thank you for your videos. It really helps from the beginner of the video which you have listed in data science playlist. 😄
The model in default method is 99.65% in train and 99.4% in test. Whereas gamma method will lower down the accuracy of the model from 99.4% to 75% therefore it has explicit shows the gamma method is unsuitable for the scenario however the regularisation has improve the train set to 1 and testing set is retained the best accuracy of model.
Yet, kernel parameter as linear has also provided a good accuracy of model.
Thank you for your guidance.
I was looking for python code to SVM... Thanks a lot... this was a great help... very clean and intuitive lecture~!
Glad it was helpful!
Thanks!!!!!!!!! for this wonderful tutorial got accuracy 99.166%
Your all concepts are so brilliant and well defined.because of these video , my concepts and doughts are now so much clear.
Glad you like them!
Thanks so much for the detailed video on SVM. This helped me a lot!
your lectures are so addictive I am enjoy learning, thank you soooooooooo much
After all possible regularisations, my highest accuracy is 99%. Thank you sir
Iradukunda, that's a pretty good score buddy. Good job 👍👌👏
What a wonderfull tutorial!! well done and well explained. Thanks a lot dude for the sharing of this expensive knowledge.
Tried couple of iterations finally I got 99.166% accuracy with all default parameters. random_state=1 while defining train test data.. Thanks a lot Sir
Nice work!
Great videos Bro, Finally understands something :)
I got score 99.4 when c=1 and gamma=scale
And i got 50 when gamma = auto
And 99.7 when gamma = auto and c=10
Thank you sir for this series. And following the tutorial with doing exercises
That’s the way to go irshad, good job working on that exercise
For digits I got highest accuracy value as 0.99 with gamma 'scale' and C=10
Thank you for your video!
That’s the way to go Коробка, good job working on that exercise
Very very good tutorial. The gentle practice of svm. Thank you
Can you make a video on title "how to determine which classification model to be used in ML according to dataset" ?
Pycaret answers that query
@@nivedhansenthilkumar964 Thanks
Look up cross validation.
@@nivedhansenthilkumar964 thanks
Yess
Can't thank you enough bro.💜🙏
Jai Shree Ram. Hope Ram bhagwaan bless your entire family.
A very solid, informative yet concise tutorial. Excellent. Please keep it up.
You are seeming to be tired from your voice but hats off your efforts !
model=SVC( C=4.0,kernel='rbf')
with this I achieved 99.25% score please,
fantastic teaching , thanks a lot!
Thank you so much for this clear and helpful explanation. well done
In linear kernel score is 96.9 percent and in rbf kernel score is 40 percent...
With gamma value the score is 0.06... And with the regularization value the score is around 45. 83 percent
When i tried iris data set with SVC default values, i got 1 accuracy. Digits data set with SVM(kernal='linear') gave 98% accuracy.
my score is 76.5 with ginni index model and 75.9 with entropy model
btw thanks for good teaching sir ji
Very good tutorial. I got 99.9% accuracy using kernel='rbf' and C=1.0.
That’s the way to go Ajeniyi, good job working on that exercise
@codebasics in the exercise solution of digits datasets, you have created a dataframe from data & target, then added target, then removed target. Does it bring any added value? What I did was just digits = load_digits(), train_test_split(digits.data, digits.target, train_size=0.8) , and do a SVM training on X train and y train.. still giving an accuracy of 98.05 (linear) 98.33 (rbf).
I intentionally did that just to show features and target variable side by side in a same table. Other than that you are right that it doesn't add any value.
@@codebasics aah okay. thanks. i tried starting machine learning many times, but never found as simple tutorials as yours. thanks a lot for the videos.
Very well-explained video. Thank you!
great tutorial.You explained all the concepts crisp and clear. liked and subbed
I calculated on digits dataset and comes with SVC = 99.16%
while with logistic regression it was = 96.38%.
So kudos to Support Vector Classification.
All your videos are just awesome❤❤❤
Thanks for your kind words of appreciation
It was really really helpful, thanks a million.
Great! Sir, Can you elaborate something about plotting the hyperplane (the decision function) in matplotlib
I want to see the best line which classifies the data
@codebasics In the exercise why target column is not generated with the code ?
df = pd.DataFrame(digits.data,digits.target)
1)What is good to have- a large gamma or regularization parameter?
2)We used only fit() but not fit_transform(), is it because the rbf Kernel will perform the transformation itself to scale the features and the target labels?
I am liking the tutorials Thanks
Glad you like them!
You teach so well...i thought i will never understand ml...
Thank you for this great series!
SVC(kernel='rbf' , C=10)
accuracy = 0.9933333333333333
(digit classifier)
thank you so much sir
i am your fan from india
guru dev ko sadar naman
great tutorial man👍👍👍
Thank You Sir, Dhaval and for the exercise I used normal rbf kernel C= 1 and got the accuracy of 0.991668
Thanks for your effort sir, but there is something I wonder.. When I fit a model, I can't see any description like you have in your jupyter notebook.(C=1, cache_size=200 etc..) I can't see them.. is there any way to see them?
Just watching the tutorial, you are not going to learn anything :-) --> We understood your intention sir. A big salute to you.
Ha ha.. nice. It is very true.
Beautiful explanation
I got 98.16% accuracy with C=2, kernel=rbf and gamma=0.001
Maximum Accuracy: 100%
Minimum Accuracy: 95 %
Avg Accuracy: 98.16%
That’s the way to go Anurag, good job working on that exercise
Great vid! but would've been nice if you had plotted the SVM line and scatter plots. Also running a few predictions would be useful.
used hyperparameter tuning here to get 100% for train and 99.72% for test...luckily data was clean cause im not very experienced in data cleaning and here i didnt even do too much data visualization
Why we are not using One hot encoding technique to transform categorical variables into binary variables in classification videos? I could have used logistic regression right?
Thankyou so much for the wonderful job!!
What an awesome tutorial.
This is great! Thank you so much for the video
Logistic Regression is giving the 100% score.....its performing better than SVC and also Decision Tree.
Hello sir, thank you so much for this video. I got 99.25% when i put C=1.
If, I use kernel='rbf' then, I got 99 % accuracy.
And for kernel='linear', I got 97.7% accuracy.
Usually we predict the values based on x_test value and we get some predictions for x_test values. And after that we compare the y_pred with the y_test, but in the video the predict code is not written or is it like that only ??
Nobody...Nobody....
does it better 😄😄😄😄
(yes you explained the best)
Thank you so much Sir! for your machine learning playlist
I am happy this was helpful to you.
OK, the excercise is cool. i got the best accuracy score using kernel = RBF.
As in this, you have given the last example for practice that same example we have solved in the Logistic regression model!! then what will be the difference b/w them?? I am talking about (Load Digits problem)
How will we get to know whether to use Logistic regression or Support vector machine?
Thank you sir for wonderful explanation.I think high regularization means simpler the model.(5.11)
For coding on 6:32, is there a way to do SVM with file reading on an CSV
Do you know a way you could look at only one data points specifically when you do the prediction at the end?
very much helpful for beigineers !! Thank you so much..
BEST DEMO ON SVM
Excellent...!!!! 😀 thanks
Roopa, thanks for the feedback
got a score of 99.16 for my test samples with C=any thing more than 2( i wonder why there were no difference between C=2 and C= 100, i got the 99.16 accuracy for all the values for C more than 2!). didn't change the gamma or the score would be destroyed! the kernel ='rbf'.
thanks for this amazing tutorial BTW! :)
Good job Amir, that’s a pretty good score. Thanks for working on exercise
why we can't write like this :- iris.target_names = df['target_names'] to add target_names in the dataframe?
You can write it that way
Lesson to be learning: Default settings work best.
Kernal = default (rbf): 0.99166666
Kernal = linear: 0.97777777
Kernal = poly: 0.9916666
Kernal = sigmoid: 0.9277777
Kernal = default, gamma = scale (default): 0.98888888
Kernal = default, gamma = auto: 0.65
Kernal = default, gamma = 100: 0.077777
Kernal = default, gamma = scale (default), C = 1(default): 0.98888888
Kernal = default, gamma = scale (default), C = 10: 0.99166666
Kernal = default, gamma = scale (default), C = 100: 0.99166666
Kernal = default, gamma = scale (default), C = 0.1: 0.95
SVC(C=4, kernel='rbf',gamma='scale')
Got an accuracy of 99.17%
hi,
you dropped for the df the target and flower name, right? why u didn't just use iris.data instead?
thanks a lot for uploading. Plz try to upload next vides soon.
And thank you sir for an awesome playlist
why we shouldn't use decision tree for this iris data set since it can also classify the points well as you shown in scatter plot. pls clarify my doubt?
well you can use decision tree as well. It was just that this tutorial was for SVM hence I used svm here. I didnt say svm is the best method to solve iris classification problem
hello great videos, loved this series. Can you please do a video on imbalanced data sets in classifications problems? Maybe just add onto a previous example you have but with a case where there are very few "1" or "true" values compared to "0" or "false" . thanks for you consideration!
Hi! Does test_train_split shuffles the matrix X?
What if we do sliding window to get dimensionality reduction (reduce the length of data), then how to classify the data if the length of data and target is different?
We have done with the iris data in Logistic Regression Exercise , which peak value was also 96 %
I got a peak value of 1 in LR
How to find linear or non-linear in the dataset if we get very large dataset ?
Exercise Results:
C=5, kernel ='rbf' & test size used 0.2, Accuracy: 99.5%
what parameters to pass in predicting model...model.predict() =..for the iris data set shown in this video
Thank you sir for this tutorial. Sir in the In[21] it is showing the error like cannot import the name 'trapezoid' from 'sklearn.utils.fixes'. Maybe it is due to the new version. What should I do?
why we have not applied feature scaling here?
**HELP**
In my jupyter notebook,
i don't get the information of modules like svc(*, c=1.0, kernel='rbf', degree=3, gamma='scale', coef0=0.0, shrinking=true, probability=false, tol=0.001, cache_size=200, class_weight=none, verbose=false, max_iter=- 1, decision_function_shape='ovr', break_ties=false, random_state=none) whenever I execute the model with X and y train datasets.
What I get is SVC(), everything blank inside.
so what should I do to get all the information????
When I did the exercise, rbf performs slightly better for me than linear. I believe when you created your notebook, the default gamma was auto. Using the scale option provides much better results than auto for rbf.
Hi. I have a aquestion, when we execute the model iteratively, the score of model is chenging. So, How can we say , what is out accuracy of model, for example. in papers as a result?,
please give me some clues, 🙏🙏🙏🙏🙏🙏🙏🙏🙏🙏
This is our 3rd Classification Algorithm...but how we choose which to use and when?
I think it should be high C corresponds to low regularization, which means the classifier don't penalize too much on classification error.
Vice versa for low C.
Sir when you didn't used the X_train and y_train parameters for checking score?
Sir cannot we do this without the library means mathematicaly using cost function and gradient
I am not getting any change in model score even after change either kernel or regularization
Please make a video on the topic "How to choose which ML algorithm for a dataset".
And thanks for amazing videos, sir.