Performance analysis

Performance graphs were already generated in the previous step. Here is how they can be regenerated from the saved DynaSigML_Model objects:

from dynasigml.dynasig_ml_model import DynaSigML_Model, load_dynasigml_model_from_file

dsml_model_hard = load_dynasigml_model_from_file("dsml_model_hard.pickle")
dsml_model_inverted = load_dynasigml_model_from_file("dsml_model_inverted.pickle")

dsml_model_hard.make_graphs('graphs_hard')
dsml_model_inverted.make_graphs('graphs_inverted')

The graphs will be located in the folder specified, which will be created if it does not already exist. For both user-specified ML models and LASSO regression, a scatter plot of predicted versus experimentally measured values is generated. A plot of predictive (testing) R-squared as a function of the beta parameter for the Dynamical Signature is also generated. Finally, for LASSO regression, a plot of predictive R-squared as a function of regularization strength is generated.

Note

If the save_testing=True flag is passed to the DynaSigML_Model constructor, text data frames are generated for all ML models with the predictions and measured values, so that the user can directly access the raw predictions if need be.