PDF Preview:
| PDF Title : | Python Machine Learning |
|---|---|
| Total Page : | 454 Pages |
| Author: | Sebastian Raschka |
| PDF Size : | 32.2 MB |
| Language : | English |
| Source : | sebastianraschka.com |
| PDF Link : | Available |
Summary
Here on this page, we have provided the latest link for Python Machine Learning PDF. Please feel free to it on your computer/mobile. For further reference, you can go to sebastianraschka.com
Python Machine Learning
we want the relative error between the numerical gradient and the analytical gradient to be as small as possible. Before we implement gradient checking, we need to discuss one more detail: what is the acceptable error threshold to the gradient check? The relative error threshold for ing the gradient check depends on the complexity of the network architecture.
As a rule of thumb, the more hidden layers we add, the larger the difference between the numerical and analytical gradient can become if backpropagation is implemented correctly. Since we have implemented a relatively simple neural network architecture in this chapter, we want to be rather strict about the threshold and define the following rules:
• Relative error <= 1e-7 means everything is okay! • Relative error <= 1e-4 means the condition is problematic, and we should look into it. • Relative error > 1e-4 means there is probably something wrong in our code.
Python Machine Learning PDF
[yasr_visitor_votes size=”medium”]