M-LOOP 3.2.0
What's new
- Allow
param_names
to be specified on the shell interface (#66). - Documentation improvements (#68, #79, #89).
- Support for visualizing Gaussian process hyperparameters when
update_hyperparameters
is false (#71). - Support bounds for Gaussian process hyperparameters (#75).
- Support updating hyperparameters for the neural net controller (#92).
- Support external controllers from config file (#97).
- Change how costs are sent between learners and controllers (#90).
- Better support for TensorFlow 2 (#100).
Bug fixes
- Visualization of Gaussian process results when using isotropic kernels (#69).
- Scaling in Gaussian process controller (#72, #74).
- Various compatibility improvements for archives (#77, #84, #90).
- Various issues with
minimum_uncertainty
(#78). - Structure of
bad_run_indexs
(#91). - Incorporate cost uncertaines into Gaussian process controller (#73).
- General visualization polish (#86, #93).