Currently, in master we have 2 possibilities for fine tuning of models in chain: tune all nodes recursively or only models in primary nodes.
However, these approaches limit the choice of tuning strategy and experiments, therefore new version (see https://github.com/nccr-itmo/FEDOT/pull/168) of fine tuning is in progress.
Instead of recursive fine tuning, the chain is transformed to "flat" structure using ChainTemplate and ModelTemplate.
For details, see class Tune.
After merge, we will have 2 versions of chain tuning, but it is planned to get rid of old version.
Therefore, this issue is devoted to:
J3FALL
commented
3 years ago
Currently, in master we have 2 possibilities for fine tuning of models in chain: tune all nodes recursively or only models in primary nodes. However, these approaches limit the choice of tuning strategy and experiments, therefore new version (see https://github.com/nccr-itmo/FEDOT/pull/168) of fine tuning is in progress. Instead of recursive fine tuning, the chain is transformed to "flat" structure using ChainTemplate and ModelTemplate. For details, see class Tune.
After merge, we will have 2 versions of chain tuning, but it is planned to get rid of old version. Therefore, this issue is devoted to: