@@Duhgy that is just to refresh some basics pertaining to Tensorflow. learning ML requires a hell lot of other steps from EDA to Feature Engineering to Feature Selection to HypterParameter Tuning.
Didn't need to watch those 2 hours video. With your video, I was able to understand the base and the rest is just research and finding codes I need. This helped so much. Thank you! You are the best!
This is the most amazing tutorial I have ever watched. I'm not ashamed to say I sometimes require extra explaining but this guy is just spot on with his explanations.
Hey guys if your trying out this video in 2023 July like me you need this line changed for it to run X = pd.get_dummies(df.drop(['Churn', 'Customer ID'], axis=1), dtype=float) the dtype=float is the most important was trying to figure why it wouldn't train all morning and just cracked it.
One thing I like about his videos is how basic he breaks down complex concepts for easy comprehension! Having knowledge is one thing but passing that knowledge on is another. Nicholas is doing great at giving that knowledge!
I've to say - great stuff, but you must be carefull with input dataset. Because of some missing values in "Total Charges", it's treated as an object instead of series of numbers. This leads to situation, when we feed layer with dimention over 6500 (which is close to cardinaltiy of training set - and this should be huge red flag - at least for example Random Forest prediction models are very bad in this circumstances). After cleaning input dataset, we end up with dimention = 45, which is reasonable in this case.
Top video, mate. Usually any Aussie who pronounces data as “day-ta” instead of the objectively superior “dah-ta” won’t win my respect. I’m willing to look past this for you xx
I am about to choose a major at university as a high school student! you would be the one who has been inspiring me to learn AI! what an amazing channel bro!
hey mate just watched your video and thought it was super useful to my learning. You explain everything very well (look good doing so) and left out the unimportant details. Thank you for this content!
@@Reeg3x 2.5 years and fairly time consuming given the various complex topics it covered (the college is nationally recognized). You'll need a heavy background in stats/computer sciences to obtain a data science job (a real one not just by name working in just excel or the like). One just can't go straight into becoming a lawyer without education just like data science.
Can you please cover fall detection ? I've been looking for tutorials on it for 6 hours on internet and i couldn't find a helpful resource.. you explain and makes things everything so easy to understand and no one does it like you !!
This was awesome man thanks. I got a good understanding of the flow of tensor flow and also the things I need to learn to become proficient. I def need to understand more about the different network types/shapes and their use cases, as well as the activation algorithms. Also is nice to know that I don't need to dive too deep into learning about the backpropagation and calculus because TF takes care of all of that!
I would like to resolve an error I came across when implementing the code: Code to train the model for a certain amount of epochs: model.fit(X_train, y_train, epochs=10, batch_size=32) Error: Failed to convert a NumPy array to a Tensor (Unsupported object type int).
i know im late but put this code before the model.fit() from numpy import asarray, float32 X_train = asarray(X_train).astype(float32) y_train = asarray(y_train).astype(float32) that should do it
When i tried to run through this exercise i ran into an issue: model.fit(X_train, y_train, epochs=200, batch_size=32) gives errorValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type int). So to get around this - I converted X and y train to float32 X_train = X_train.astype('float32') y_train = y_train.astype('float32') Later i ran into a similar issue with: y_hat = model.predict(X_test) y_hat = [0 if val < 0.5 else 1 for val in y_hat] So again - converted X_test = X_test.astype('float32') Everything seemed to complete as expected with 0.79 accuracy score. Thoughts?
This video is awesome, I have two questions because I'm new in Tensorflow, 1- Do we need to encode numeric data in the data sheet before we start building the model?, because I didn't see that in the video. 2- How we can map the prediction results 0, 1 to Yes, No as per the data sheet?
Customer churn is the percentage of customers who stopped purchasing your business's products or services during a certain period of time. Your customer churn rate indicates how many of your existing customers are not likely to make another purchase from your business.
I had trouble understanding that as well, although in fairness to Nicholas, I think his real purpose was to show the process of TF neural network synthesis, as opposed to a real use case of one shot encoding of the columns. I dropped Monthly and Total Charges (and tenure as well) as I did not see any benefit of adding so many columns. Perhaps that was a vestige of an earlier video? Still pretty damn good for 10 minutes...
How did you decide number of neurons to include in your sense layers? Do these relate to the number of feature columns in your data set at all? Or just a random/empiric choice?
This is really cool and I know this tutorial is a bit old at this stage but just a question on how you would run inference on the saved model with new input data?
Lately I've been developing a large Tensorflow model, and I'm getting out of memory errors, from what I've learned it seems the best solution to this road block is gradient checkpointing, however there is little to no resources online about it. Could you make a video covering gradient checkpointing?
is there any rule like, to load the model, the model should be saved in same computer and with same version of tensorflow. I am asking this because, i downloaded a pretrained model from tensorflow zoo into my pc. Then I use this load function with model folder. its not working
QQ: why do you use sequential if the TensorFlow docs say it shouldn't be used if "your model has multiple inputs or multiple outputs"? I saw your csv has a ton of factors and you have one output you are looking for. I have a very similar project where I have nfl game stats and I need to predict if the home team won. Any help is appreciated!
Hi Nick, can you comment which is the business case with this model? What do we want to predict? In which escenario can do we use this model? We got the model and the accuracy_score but then how can we use it? Quite new to Data Science. Thanks.
Heya Juan, so churn prediction is a really popular ML task in businesses because it's to do with predicting customers that might leave our company. This is important to get on top of (and ideally try to keep the customer) because it costs a lot more to attract a new customer than to keep an old one!
Please help. I ran the tutorial in google colab, got the model out to drive, then back into the colab notebook. I dont understand what i am supposed to do with the model once it's ready. This tutorial doesnt like, open it up and look at what it learned. Can someone please offer guidance?
What does the output "accuracy score" (of ~59%) actually mean on the Churn model? Does it mean there is a correlation between one or more columns from the features and the target variable?
Does it enable to train any kind of task required to achieve? can it learn from it for example how to do videos correctly? (Im a complete total noob in AI so I have no idea)
I don't understand these 2 lines from section 0 Import data: ➡ X = pd.get_dummies (df.drop(['Churn', 'Customer ID'], axis=1)) ➡ X_train, X_test, y_train, y_test = train_test_split (X, y, test_size=.2) What's the difference between X_train and X_test?
I love your videos! I have a small problem with this one though. This is rather keras and not tensorflow. With plain tensorflow you need lots more coding (which of course comes with greater flexibility)
Heya, ultimately using Keras with a Tensorflow backend. You still have a lot of flexibility running using the Sequential API, I'd agree though, there is a lot more flexibility using direct tensorflow layers. In my opionion however unless you're creating complex models or performing research it seems like overkill for most use cases.
Sir i am gotting lot of error in tensorflow pkg importing where i got a complete tensorflow pkg or pre installed pkg of tensorflow and its depwndencies
Hi Nicholas great tutorial, I came to youtube after struggling to understand many of the tensorflow tutorials, Im not really following how the testing process is being done I see we have a very large csv but what is the criteria for whether the test was accurate or not
Hey Nicholas great video thanks for this. Trying to deploy this on AI platform,GCP. There I would need to pass on one sample, but kinda getting stuck how would that be one-hot encoded as we wot be passing the complete dataset.
What is the output of this ? Having a number like 0.8 is of no use when I want to see how many have churned. You could just have put a filter on the Excel sheet on the Churn column !
I have a question. I have heard that for deep learning models or for computer vision we use Tensorflow or Pytorch. So my question is that please explain to me whether these are some applications like visual studio code or are these some kind of add ons or extensions (packages to install that work on different applications)? Actually I am new in the field. Also what Jupyter Notebook, How is it related to these ?
Tensorflow and Pytorch are just "libraries" to work with DL and to build models and so on Python itself is just the language and it has alot of libraries to use in many other different fields You can write python in a file *.py then run it normally but jupyter notebooks provides easier way to run python code where you can write cells and run them independently so that you can run any part of the code as you want and it's widely used in ML and data science where we tend to do alot of trial and error so we don't need to run the whole code over and over
These videos are so good. A whole end-to-end project in 10 minutes. And a bit of humour and art tossed in there.
Thanks so much @Shivan! Glad you enjoyed it!
#TensorFlow-- python Library #Explanation with Example
ua-cam.com/video/ojevo88RVaE/v-deo.html
Yeah but you aren’t taught anything, you cant learn ml in 10 mins I’m sorry
@@Duhgy that is just to refresh some basics pertaining to Tensorflow. learning ML requires a hell lot of other steps from EDA to Feature Engineering to Feature Selection to HypterParameter Tuning.
@@NicholasRenotte ❤
I normally never comment on tutorial videos but this was very excellently done! This was exceedingly concise and clear
I like how he is doing 10min tutorial but still included a humor intro
😆 gotta try to stay a little funny!
Didn't need to watch those 2 hours video. With your video, I was able to understand the base and the rest is just research and finding codes I need. This helped so much. Thank you! You are the best!
YESSS! Once you get the structuring it's all just a matter of building different architectures where needed!
This is the most amazing tutorial I have ever watched. I'm not ashamed to say I sometimes require extra explaining but this guy is just spot on with his explanations.
Hey guys if your trying out this video in 2023 July like me you need this line changed for it to run X = pd.get_dummies(df.drop(['Churn', 'Customer ID'], axis=1), dtype=float)
the dtype=float is the most important was trying to figure why it wouldn't train all morning and just cracked it.
very helpful thanks!
Thank you so much !!
Thanks for reaching the heart of the matter (4:07) so quickly and then explaining these '4 lines' so well.
Stoped you enjoyed it @Saptadeep!
This tutorial is an absolute life-saver. Well done!
One thing I like about his videos is how basic he breaks down complex concepts for easy comprehension!
Having knowledge is one thing but passing that knowledge on is another. Nicholas is doing great at giving that knowledge!
Absolutely brilliant. End-to-end in just 10 minutes. Very explicit. Thanks for sharing
Went along with you and got .8 on the last epoch, but had .78 on the accuracy score. Loved this tutorial; it was so well explained. Thanks!
Awesome stuff @Tia, awesome work!
I've to say - great stuff, but you must be carefull with input dataset.
Because of some missing values in "Total Charges", it's treated as an object instead of series of numbers.
This leads to situation, when we feed layer with dimention over 6500 (which is close to cardinaltiy of training set - and this should be huge red flag - at least for example Random Forest prediction models are very bad in this circumstances).
After cleaning input dataset, we end up with dimention = 45, which is reasonable in this case.
#TensorFlow-- python Library #Explanation with Example
ua-cam.com/video/ojevo88RVaE/v-deo.html
Although this isn't an actual tutorial, it is cool to see you build a model so quickly!
This is a perfect introduction to sharing with people on any team that works with someone working with ML. :D
Sir, as of all your other tutorials, it is so self-explanatory and clearly defined. Thank you so much.
#TensorFlow-- python Library #Explanation with Example
ua-cam.com/video/ojevo88RVaE/v-deo.html
epitome of greatness. Fashion Model looking guy having the best ML/AI lessons on youtube! xD
Finally a concise introduction to TensorFlow.
Top video, mate. Usually any Aussie who pronounces data as “day-ta” instead of the objectively superior “dah-ta” won’t win my respect. I’m willing to look past this for you xx
Cheers @Billy, I'll drop a "dah-ta" for you in one of the future videos 🤣my US colleagues have given up on trying to convert me!
Thankyou so much Nicholas, this is what I was looking for, whole story in 10 minutes, Tq so much,brilliant effort.
It's a bit of a crash course but it goes through the basics right?! 😃
Thank you! Up until now I've only done old skool supervised learning. With your tutorial I'm ready to enter the big leagues!!
Excellent presentation. Straight to the point, easy to follow and well explained.
I am about to choose a major at university as a high school student! you would be the one who has been inspiring me to learn AI! what an amazing channel bro!
YESSS, go getem!
Short and sweet! I'll add it to my memory palace. Thanks again.
Exactly what I am looking for. Something straight forward
hey mate just watched your video and thought it was super useful to my learning. You explain everything very well (look good doing so) and left out the unimportant details. Thank you for this content!
A much needed video! Thank you for the great work!!
Understood one word out of 5, but this will for sure make me wants to work with it.
Excellent video, this was short, very clear, and easy follow. Great job, and thank you for this!
I totally love your dedication on each videos bro !! Thanks for your hard work and keep it going !!!
#TensorFlow-- python Library #Explanation with Example
ua-cam.com/video/ojevo88RVaE/v-deo.html
Thanks a mil for checking it out @Nguyen Long!!
Thanks for sharing this! Can’t wait to watch some more of your content.
Thanks so much @Scarlett, plenty more to come!
This is really useful. Give me a much clearer idea on how it works.
I remember my grad days for data science and this would still scare me for a test like that lol. Great video!
Hahahah, ikr, man I've been working with TimeDistributed layers right now and it's giving me the same nightmares!
How long and time consuming was grad school for data science? Could it be done with a full time data science job?
@@Reeg3x 2.5 years and fairly time consuming given the various complex topics it covered (the college is nationally recognized). You'll need a heavy background in stats/computer sciences to obtain a data science job (a real one not just by name working in just excel or the like). One just can't go straight into becoming a lawyer without education just like data science.
@@protovici1476I’ll let you know if that last part is true or not after my interview next week.
Add to it 1 month to start understanding what it is that you are doing and how to improve your models.
I have been "tensored"! Hopefully this is the beginning of my AI career! Thank you
You are the best, Nicholas. Just Brilliant!!
Thanx for sharing your knowledge with us bro. U explain so easily and effectively
Finally I understood thank you for explicit expansion of those attributes.
This one video made me subscrib immediately to your channel and like all videos of you ! so perfect and time saving
Can you please cover fall detection ?
I've been looking for tutorials on it for 6 hours on internet and i couldn't find a helpful resource..
you explain and makes things everything so easy to understand and no one does it like you !!
Perfect introduction, thanks a lot! Got me over that first big hurdle
Excellent video! Concise and every step well-explained throughout.
This was awesome man thanks. I got a good understanding of the flow of tensor flow and also the things I need to learn to become proficient. I def need to understand more about the different network types/shapes and their use cases, as well as the activation algorithms. Also is nice to know that I don't need to dive too deep into learning about the backpropagation and calculus because TF takes care of all of that!
your videos are really helping man! appreciate it!
Okay! This was much needed. Thenks!
Thanks so much @fukat!!
I would like to resolve an error I came across when implementing the code:
Code to train the model for a certain amount of epochs:
model.fit(X_train, y_train, epochs=10, batch_size=32)
Error:
Failed to convert a NumPy array to a Tensor (Unsupported object type int).
same
@@PongsatornKanjanasantisak same what do i do
i know im late but
put this code before the model.fit()
from numpy import asarray, float32
X_train = asarray(X_train).astype(float32)
y_train = asarray(y_train).astype(float32)
that should do it
Seriously killer teaching (I assume to be Australian) sir.
When i tried to run through this exercise i ran into an issue:
model.fit(X_train, y_train, epochs=200, batch_size=32)
gives errorValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type int).
So to get around this - I converted X and y train to float32
X_train = X_train.astype('float32')
y_train = y_train.astype('float32')
Later i ran into a similar issue with:
y_hat = model.predict(X_test)
y_hat = [0 if val < 0.5 else 1 for val in y_hat]
So again - converted X_test = X_test.astype('float32')
Everything seemed to complete as expected with 0.79 accuracy score.
Thoughts?
X = pd.get_dummies(df.drop(['Churn', 'Customer ID'], axis=1), dtype=float)
OK. you have the biggest eyes on the planet. YOU WIN!
Dude, very nice compact video, thanks for making it!
So perfectly explained. Thank you
So glad you enjoyed it @Anirban!
so brilliantly!!!
Very unselfish 🙏💐
This video is awesome, I have two questions because I'm new in Tensorflow,
1- Do we need to encode numeric data in the data sheet before we start building the model?, because I didn't see that in the video.
2- How we can map the prediction results 0, 1 to Yes, No as per the data sheet?
You are awesome. This was so helpful. Thank you!
EXCELLENT tutorial.
Thank you..
Awesome stuff @Mohamed, glad you enjoyed it!
you didnt explain what "churn" means😓😓😓
Customer churn is the percentage of customers who stopped purchasing your business's products or services during a certain period of time. Your customer churn rate indicates how many of your existing customers are not likely to make another purchase from your business.
He's still learning 😅
@@amleth_prince_of_denmark thx!!
This is brilliant! Well done Nicholas so helpful!
Thanks so much @Kevin! If you're interested, maybe check out Object Detection with Tensorflow next! ua-cam.com/video/yqkISICHH-U/v-deo.html
Watch it in 2x to learn Tensorflow in 5 minutes
This guy tensorflows
Really great Churn Model explained in TensorFlow but,
why use pd.get_dummies() for the data preprocessing?
I had trouble understanding that as well, although in fairness to Nicholas, I think his real purpose was to show the process of TF neural network synthesis, as opposed to a real use case of one shot encoding of the columns. I dropped Monthly and Total Charges (and tenure as well) as I did not see any benefit of adding so many columns. Perhaps that was a vestige of an earlier video? Still pretty damn good for 10 minutes...
I could implement a churn model in rt thanks to Nicholas
Great video, Nicholas.
How did you decide number of neurons to include in your sense layers? Do these relate to the number of feature columns in your data set at all? Or just a random/empiric choice?
This was very helpful. Thank you very much.
This is really cool and I know this tutorial is a bit old at this stage but just a question on how you would run inference on the saved model with new input data?
very helpful. made this seem "easy", which it def is not. Thanks!
great job and nice and concise work..
Another great video 👏
Yeayyyahhh, thanks @Gustavo!!
Lately I've been developing a large Tensorflow model, and I'm getting out of memory errors, from what I've learned it seems the best solution to this road block is gradient checkpointing, however there is little to no resources online about it. Could you make a video covering gradient checkpointing?
Cheers mate u are doing brilliant things!
amazing video easy to understand
is there any rule like, to load the model, the model should be saved in same computer and with same version of tensorflow. I am asking this because, i downloaded a pretrained model from tensorflow zoo into my pc. Then I use this load function with model folder. its not working
QQ: why do you use sequential if the TensorFlow docs say it shouldn't be used if "your model has multiple inputs or multiple outputs"? I saw your csv has a ton of factors and you have one output you are looking for. I have a very similar project where I have nfl game stats and I need to predict if the home team won. Any help is appreciated!
Hi Nick, can you comment which is the business case with this model? What do we want to predict? In which escenario can do we use this model? We got the model and the accuracy_score but then how can we use it? Quite new to Data Science. Thanks.
Heya Juan, so churn prediction is a really popular ML task in businesses because it's to do with predicting customers that might leave our company. This is important to get on top of (and ideally try to keep the customer) because it costs a lot more to attract a new customer than to keep an old one!
To the mark. Keep going!
Please help. I ran the tutorial in google colab, got the model out to drive, then back into the colab notebook.
I dont understand what i am supposed to do with the model once it's ready.
This tutorial doesnt like, open it up and look at what it learned.
Can someone please offer guidance?
What does the output "accuracy score" (of ~59%) actually mean on the Churn model? Does it mean there is a correlation between one or more columns from the features and the target variable?
Subbed! The video is so good!
0:23 _NICHOLAS RENOTTE - WORRIED ABOUT THE TIME LIMIT_ *Talking fast* LOL, that made me laugh really hard. I also enjoyed the video.
awesome and simple explanation
What is Customer Churn?
Heya, it's to do with predicting customers that are likely to leave your business (e.g. go to another company or stop using your service altogether)!
Yeh,It's Really helpful ,Thank you for the video
Great video, veeery instructive and useful. ¡Gracias!
Does it enable to train any kind of task required to achieve? can it learn from it for example how to do videos correctly? (Im a complete total noob in AI so I have no idea)
can u tell me why we are using X and Y variable , and what does the get_dummies func do .
Fantastic, after watching this video, making a couple of notes, I'm off to apply for an AI job at NASA.
9:18 you used the sigmoid activation function that outputs 1 or 0
Why the need for the if statement
I don't understand these 2 lines from section 0 Import data:
➡ X = pd.get_dummies (df.drop(['Churn', 'Customer ID'], axis=1))
➡ X_train, X_test, y_train, y_test = train_test_split (X, y, test_size=.2)
What's the difference between X_train and X_test?
I love your videos! I have a small problem with this one though. This is rather keras and not tensorflow. With plain tensorflow you need lots more coding (which of course comes with greater flexibility)
Heya, ultimately using Keras with a Tensorflow backend. You still have a lot of flexibility running using the Sequential API, I'd agree though, there is a lot more flexibility using direct tensorflow layers. In my opionion however unless you're creating complex models or performing research it seems like overkill for most use cases.
Great video..
Only thing that troubled me was the data selection using pandas but I will find out
Awesome! Want to share? Happy to help out!
Why 32 and 64 units in the dense layers? How to know the no. of neurons to have in my NN layers?
Thank you for the awesome video.
How to fix AttributeError: module 'numpy' has no attribute 'object' while importing tensorflow?
Sir i am gotting lot of error in tensorflow pkg importing where i got a complete tensorflow pkg or pre installed pkg of tensorflow and its depwndencies
Hi Nicholas great tutorial, I came to youtube after struggling to understand many of the tensorflow tutorials, Im not really following how the testing process is being done I see we have a very large csv but what is the criteria for whether the test was accurate or not
Ppp
Pplppp
wow great and fast ! thank you!
Can you use ordinal data in a predictive model?
please do a video on how to improve the accuracy
absolutly subscribed
If you added more dense layers would the model be able to predict better?
Hey Nicholas great video thanks for this. Trying to deploy this on AI platform,GCP.
There I would need to pass on one sample, but kinda getting stuck how would that be one-hot encoded as we wot be passing the complete dataset.
Not familiar with GCP but you could access a single row from the encoded dataframe by using X_train.values[0]
#TensorFlow-- python Library #Explanation with Example
ua-cam.com/video/ojevo88RVaE/v-deo.html
What is the output of this ? Having a number like 0.8 is of no use when I want to see how many have churned. You could just have put a filter on the Excel sheet on the Churn column !
Sir I am getting this error after model.fit, ValueError: name for name_scope must be a string. Can you please tell me how to solve it
Have you set a name for something, looks like the name should be set to a string?
I have a question. I have heard that for deep learning models or for computer vision we use Tensorflow or Pytorch. So my question is that please explain to me whether these are some applications like visual studio code or are these some kind of add ons or extensions (packages to install that work on different applications)? Actually I am new in the field. Also what Jupyter Notebook, How is it related to these ?
Tensorflow and Pytorch are just "libraries" to work with DL and to build models and so on
Python itself is just the language and it has alot of libraries to use in many other different fields
You can write python in a file *.py then run it normally but jupyter notebooks provides easier way to run python code where you can write cells and run them independently so that you can run any part of the code as you want and it's widely used in ML and data science where we tend to do alot of trial and error so we don't need to run the whole code over and over