Parameters & Cost log per iteration

Course Q&A Deep Learning Specialization Neural Networks and Deep Learning
1

Hello, at the moment Im doing course C1 week2 Lab03 Feature Scaling and Learning Rate.
I found super useful to run the gradient descent function and observe how values change for the Cost function and for the parameters w in each iteration. This historical information per each iteration I believe provide insightful information about if gradient descent is converging: if Cost is close to 0 and if parameters Ws values are not changing from previous iteration.

My question is: is possible to get above historical info from sklearn model SGDRegressor? How I can get from SGDRegressor Ws and Cost values per each interaction?
Many thanks

2

I do not think sklearn provides such an interface. (as far as I know…)

You can use SGD in Tensorflow, and get history from model.fit for visualization as you wrote.

image
image1575×563 71.6 KB

Tensorflow provides more useful tool, Tensorboard.
You can easily pass logs to Tensorboard for visualization and analysis. Here is an example.

image
image2592×1676 358 KB

3

thanks Nobu I will check Tensorboard sound a good tool