Conditional Neural Processes

[mastoffel]

• implements a CNP a la Garnelo et al;
• tensorflow code here
• neural processes overview here

CNP's do meta learning, they train on many datasets with context_x, context_y, target_x as inputs, not one dataset with X, y. This PR makes it work anyway. We create a meta-dataset from standard X, y input, with variable context sizes for each episode. This is then fed into the CNP for training. The full training dataset is stored as an attribute and provided as the context during prediction, so that we can do cnp.predict(X) in the same way as for the other models.

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
autoemulate/emulators
__init__.py
conditional_neural_process.py					141-142, 246-247, 253-281, 286, 300
neural_net_torch.py
autoemulate/emulators/neural_networks
base.py
cnp_module.py
datasets.py					11
losses.py
tests
test_cnp.py
test_cnp_dataset.py
test_estimators.py
Project Total

This report was generated by python-coverage-comment-action

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 97.72727% with 9 lines in your changes missing coverage. Please review.

Project coverage is 91.57%. Comparing base (8c482e6) to head (4659143).
Report is 4 commits behind head on main.

Files	Patch %	Lines
...utoemulate/emulators/conditional_neural_process.py	92.00%	8 Missing ⚠️
autoemulate/emulators/neural_networks/datasets.py	97.95%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #217      +/-   ##
==========================================
+ Coverage   91.14%   91.57%   +0.43%     
==========================================
  Files          45       51       +6     
  Lines        2393     2802     +409     
==========================================
+ Hits         2181     2566     +385     
- Misses        212      236      +24     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.