Exploring options for training neural network classifiers
You have a lot of options when training neural network models with TPOT. The whole neural network story is still new and experimental with TPOT, requiring a bit more manual work than regular scikit-learn estimators.
By default, TPOT won't use the neural network models unless you explicitly specify that it has to. This specification is done by selecting an adequate configuration dictionary that includes one or more neural network estimators (you can also write these manually).
The more convenient option is to import configuration dictionaries from the tpot/config/classifier_nn.py file. This file contains two PyTorch classifier configurations, as visible in the following diagram:
Figure 7.8 – TPOT PyTorch classifier configurations
From the preceding diagram, you can see that TPOT can currently handle two different types of classifiers based on deep learning libraries:
- Logistic...
