matthewsholden / skillassessment Goto Github PK

In many cases, if the workflow steps are identified, it is not necessary (and perhaps detrimental) to use the overall "MetricValue" for the entire procedure in addition to the task-specific metrics.

For these cases, we should allow the overall "MetricValue" to be ignored, and allow computation to proceed on only the task-specific metrics.

Currently, any table that does not have the same order of metrics as the "test" table is ignored. But sometimes, tables have different orders of metrics depending on how the scene was setup. Skill assessment on tables with different ordering should be permitted.

The description should include:

1. An outline of how the method works.
2. Details on exactly how the computation worked for the specific case.

Add an option allowing the scores for the metric-task to be displayed instead of the metric-task values. These scores are often more informative, because they allow the trainee to understand which aspects of which tasks they need to improve in.

In addition to the linear combination method. This uses the ground-truth skill classes to find an optimal linear combination.

At present, the metric membership functions are assumed to be Gaussian. But in practice, metrics do not follow a Gaussian distribution.

Instead, the membership functions could be modelled non-parametrically using kernel density estimation methods.

The full metric names (e.g. Elapsed Time [Any = StylusToReference]) are not necessarily intuitive to all users. We should add a mechanism that allows these names to be translated into human-readable metrics, such that the feedback is more understandable to users.

The log is being cluttered with too much text during skill assessment. Only critical information should be reported; other situations should be handled silently.