Mr. Sri, I am astounded and in utter awe how you are able to learn, digest, create and then teach cutting edge features in such detail. I know I speak for most of your audience when I say this. Can you maybe do a video of a completely different kind? Starting with the WHY - why do you do what you so. WHAT - What is your drive - what is it that motivates you. HOW - how do you manage your day job, family and personal time in 24 hours or 1 week to be so prolific. Big ask but I know there are thousands if not millions wanting to know this. :) TIA! -Praveen P
Hi Praveen.. Thank you for your kind words. Second part of what you asked looks more self glorifying and kind of want to stay away for it .. :) I have just re-purposed my time that I used to spend on other activities like watching movies into this
Hi Sir. Did you realize that the tuner.search is only training using 2 epochs ? You can increase that number of epochs as well as remove the EarlyStopping callback that the training will keep training only for two epochs.
@@AIEngineeringLife Thanks for your response, I know keras tuner is model agnostic, it works with sklearn algorithms as well so wanted to see how it is used to tune those models. If it in not a priority at this time, that is acceptable
@@opalkabert Yes it does with scikit learn and will do it in few weeks.. Meanwhile you can check my playlist below in specific Autosklearn that works on top of scikit learn ruclips.net/p/PL3N9eeOlCrP5HAtwktnf5Q_ut9t_80hUm It has few other framework as well
Nice tutorial! While watching I came up with some questions regarding overfitting/generalization: 1. Does Keras-Tuner searches for the best model considering overfitting? We specify the parameters for training (epochs & batch size), so is Keras-Tuner somehow considering overfitting in the model comparison or is it just comparing the acc of each model after the specified epchos rigardingless the number of epoch leads to overfitting or not? 2. If it does not, is the tuner still usefull? 3. If it does, can we show the number of epochs used for each model in the model report?
David, You can set early stopping parameters so model does not overfit and it validates it against validation accuracy. It can be a hyperparameter Yes you can get each model in the output and epoch it was used to train for and The way I use it get quick intuition on right architecture that works for the data and then fine tune it based on need
Hi Sir, I really like the way you explain the topics via notebooks by going over each block of code. By seeing this video 2 questions came to my mind which are as follows : - 1. Can this video be considered as an extension to the AutoML series you uploaded previously? (I went through H2O ML, TPOT, Auto ViZMl and Autosklearn from your videos.) 2. Can you help me get some clarity on the differences between the Keras Tuner and other AutoML packages( such as Auto Keras and etc)? Thanks
Yes Vipul it is extension to AutoML but more specifically for Neural Network. Earlier ones were good for structured data To my knowledge Auto Keras uses Keras Tuner in backend. Unless something has changed very recently and they have their own Tuner
@@AIEngineeringLife Thank you so much sir, for the quick reply... keep up the good work, I have shared your channel with my friends too, we like your content a lot.
I got an error when working with pretrained model. Trying to get optimal learning rate. AttributeError: 'HyperParameters' object has no attribute 'get_shape'
Sir! I am following your channel since a month or so! One thing that daunts me is not knowing Deep Learning, I know many of my friends who are learning DL but ignoring the detailing of ML.So should one properly learn ML first and then start with DL or proceed with both? I know there are no hard dependencies between them but still asking you, being an expert would help!
Anand it is a matter of choice but if I would learn I will understand ML foundations first. DL becomes lot easier and also remember 80% of projects today are tabular where ML works better and is explainable
If you are talking about some breaks in voice then my initial videos had some issue but rectified it in current video recording. Sorry for inconvenience with this one
Mr. Sri, I am astounded and in utter awe how you are able to learn, digest, create and then teach cutting edge features in such detail. I know I speak for most of your audience when I say this. Can you maybe do a video of a completely different kind? Starting with the WHY - why do you do what you so. WHAT - What is your drive - what is it that motivates you. HOW - how do you manage your day job, family and personal time in 24 hours or 1 week to be so prolific.
Big ask but I know there are thousands if not millions wanting to know this. :)
TIA!
-Praveen P
Hi Praveen.. Thank you for your kind words. Second part of what you asked looks more self glorifying and kind of want to stay away for it .. :)
I have just re-purposed my time that I used to spend on other activities like watching movies into this
@@AIEngineeringLife and of course there is humility as well 😋
Brilliant tutorial. Thank you so much.
Awesome Tutorial Sir, Thank you so much!
What is the best way to summarize the tuner.result_summary() results to a table format for reporting all hyperparameter tuning results.
Hi Sir. Did you realize that the tuner.search is only training using 2 epochs ? You can increase that number of epochs as well as remove the EarlyStopping callback that the training will keep training only for two epochs.
Good job sharing this. Could you do another one showing how to us this to tune tf.keras model using a tabular data?
For tabular data I have videos on h2o automl, tpot, autoviml and others. While keras Turner is possible never thought many use NN for tabular data
@@AIEngineeringLife Thanks for your response, I know keras tuner is model agnostic, it works with sklearn algorithms as well so wanted to see how it is used to tune those models. If it in not a priority at this time, that is acceptable
@@opalkabert Yes it does with scikit learn and will do it in few weeks.. Meanwhile you can check my playlist below in specific Autosklearn that works on top of scikit learn
ruclips.net/p/PL3N9eeOlCrP5HAtwktnf5Q_ut9t_80hUm
It has few other framework as well
Great video. Can you please share the notebook you've used in this video for our reference?
Here you go Tarun.. github.com/srivatsan88/RUclipsLI/blob/master/Tensorflow_Keras_Tuner_Demo.ipynb
Nice tutorial!
While watching I came up with some questions regarding overfitting/generalization:
1. Does Keras-Tuner searches for the best model considering overfitting? We specify the parameters for training (epochs & batch size), so is Keras-Tuner somehow considering overfitting in the model comparison or is it just comparing the acc of each model after the specified epchos rigardingless the number of epoch leads to overfitting or not?
2. If it does not, is the tuner still usefull?
3. If it does, can we show the number of epochs used for each model in the model report?
David, You can set early stopping parameters so model does not overfit and it validates it against validation accuracy. It can be a hyperparameter
Yes you can get each model in the output and epoch it was used to train for and The way I use it get quick intuition on right architecture that works for the data and then fine tune it based on need
Hi Sir,
I really like the way you explain the topics via notebooks by going over each block of code.
By seeing this video 2 questions came to my mind which are as follows : -
1. Can this video be considered as an extension to the AutoML series you uploaded previously?
(I went through H2O ML, TPOT, Auto ViZMl and Autosklearn from your videos.)
2. Can you help me get some clarity on the differences between the Keras Tuner and other AutoML packages( such as Auto Keras and etc)?
Thanks
Yes Vipul it is extension to AutoML but more specifically for Neural Network. Earlier ones were good for structured data
To my knowledge Auto Keras uses Keras Tuner in backend. Unless something has changed very recently and they have their own Tuner
@@AIEngineeringLife Thank you so much sir, for the quick reply... keep up the good work, I have shared your channel with my friends too, we like your content a lot.
I got an error when working with pretrained model. Trying to get optimal learning rate.
AttributeError: 'HyperParameters' object has no attribute 'get_shape'
Rajnish can you please paste that part of the code?
@@AIEngineeringLife Here is my notebook:- www.kaggle.com/lucca9211/fork-of-flowers-classification-on-tpu
with complete error info
Sir! I am following your channel since a month or so! One thing that daunts me is not knowing Deep Learning, I know many of my friends who are learning DL but ignoring the detailing of ML.So should one properly learn ML first and then start with DL or proceed with both? I know there are no hard dependencies between them but still asking you, being an expert would help!
Anand it is a matter of choice but if I would learn I will understand ML foundations first. DL becomes lot easier and also remember 80% of projects today are tabular where ML works better and is explainable
Hi sir the video is beneficial for me.
Do some on PDF parsing in python.
Will try Pavan on pdf..
Thank you Sir.
good one. voice may have some problem
If you are talking about some breaks in voice then my initial videos had some issue but rectified it in current video recording. Sorry for inconvenience with this one