Optimizers Losses And Metrics - Keras
Here I go over the nitty-gritty parts of models, including the optimizers, the losses and the metrics. I first go over the usage of optimizers. Optimizers are essentially used to train models in very specific ways - in order to figure out how to train models, check out the list of available optimizers (using keras.optimizers) and find the optimizers you want, after which you may specify lots of hyper parameters for it and then pass it in.
I then talk about Parameters common to all Keras optimizers. All optimizers have a couple of common parameters, these are generally important and good to know regardless of what type of learning you're trying to do. Here I first walk you through learning rate (which is basically how fast the model trains and learns), then I explain clip norm (which is the max speed at which the model can learn at a particular step).
After this, I explain the usage of loss functions: These functions need to be minimized - I go through some examples. What is important to keep in mind is that loss functions are distance metrics between the true value and the predicted value and you must always try to minimize it!
Finally I talk about the usage of metrics: Any loss function can be a metric. A metric is basically a function that is used to judge the performance of your model. I also walk you through the process of making your own metric.
Links:
1) Update video: • 01. Update
2) Scikit Learn Intro video: • Intro to Scikit Learn
3) Link to the History of Deep Learning video will be up soon!
4) The Hitchhiker's Guide to Python - one of the best handbooks to the installation, configuration, and usage of Python that I have come across: http://docs.python-guide.org/en/latest/
5) Link to Keras: https://keras.io
6) Link to TensorFlow: https://www.tensorflow.org
7) GitHub link to a-bit-of-deep-learning-and-keras notebooks! - https://github.com/knathanieltucker/a...
Watch video Optimizers Losses And Metrics - Keras online without registration, duration hours minute second in high quality. This video was added by user Data Talks 18 November 2017, don't forget to share it with your friends and acquaintances, it has been viewed on our site 12,056 once and liked it 101 people.