Neural Networks Demystified [Part 7: Overfitting, Testing, and Regularization]

We've built and trained our neural network, but before we celebrate, we must be sure that our model is representative of the real world.

Supporting Code:

In this series, we will build and train a complete Artificial Neural Network in python. New videos every other friday.

Part 1: Data + Architecture
Part 2: Forward Propagation
Part 3: Gradient Descent
Part 4: Backpropagation
Part 5: Numerical Gradient Checking
Part 6: Training
Part 7: Overfitting, Testing, and Regularization

@stephencwelch

[venim]

So according to your less overfitted results I can now:
- Study for 1 second
- Sleep for 9 hours
to get 84/100 ! Amazing!!!

[Doctorcanniball]

That was one of the greatest video series i watched on youtube in 7 years.

[channagirijagadish]

This introduction probably is the finest introduction ever comparable to a good Feynman's lecture in physics. It reduces the subject matter to its very foundation so that anybody can understand and jump to more advanced topics.

[imoereddd]

Videos like this can only be made by people who really understand what their doing and you sir, know exactly what you're doing! Thank you.

[kylemckenzie]

For those getting "ValueError: setting an array element with a sequence." on Python 3.4.x, it's happening because (sum(self.W1**2) and (sum(self.W2**2) are returning 1x3 arrays and we need a single float from each. If you change J in costFunction(self, X, y) to
"J = + \

you will have the right dimensions and NN will work as expected passing the test.

Hope that helps someone!

[zhuyunhua]

I am an experimental biologist, I have struggled to learn math, python and machine learning for half a year, and this video just connected everything I have to an unified picture. Thank you very much!

[andrewbeaven]

From an experienced the best exposition of a modelling technique I have seen.

[shomodor]

Thank you for this series ! 
Feedback : Visual elements in your videos, ( like : "radial knob control" to show weight adjustment in one of your previous videos) are the little things that really made me deeply understand what exactly is happening. It gives a strong intuition to the underlying process. Waiting for future videos !

[geneus]

This is the best Neural Networks tutorial I find on Youtube !

[eeveeta]

Thank you, thank you, thank you. I'm binge watching your videos on a Friday night and I'm smiling like an idiot, I can't imagine a better way to spend my evening.

[barnumcastillo]

I cannot express how much this series of videos helped me. Learning the math notation is great but knowing also the implementation in matrices is a must, I used your explanation to implement the vanilla flavor of the perceptron multilayer.

Thank you

[Youtuber__]

This video series is insane, the amount of detail explored and not even using many machine learning packages. Develops from scratch. Really awesome job!

[photinoman]

After a poorly prepared ML workshop in the morning, I decided to watch a youtube video. You, sir, deserve a medal! I just need to sit down, redo all the math and write the code again! Thank you for the videos.

[HughMcBrideDonegalFlyer]

This series is probably the best I have seen on the the web bar none,

[adrijusremeskevicius]

Even though I know nothing about Python language, this series of videos is the best explanation of how feedforward neural network works. I wish I could find something similar on recurrent neural net called NARX for time-series problems.

[Eyaeyaho]

The best YouTube tutorial videos ever by far.

[beaconofwierd]

Without a doubt the best tutorial/explanation I've seen on youtube, ever, in all categories :D

[krlospatrick]

That was one of the greatest video series about NN I've watched.
Thanks a lot!

[goldfishlaser]

This was a great series to give myself a recap of neural networks after a small break from study.

[texheels]

I am not a data scientist, and found that series very educational. Biggest benefit for me was the "schematic approach" to machine learning in the first video and they key "fundamental gotchas" (such as overfitting) to look out for along the process of executing machine learning. Tagging these series for future referencing....thank you so much for posting.