dieterich-lab / asyh Goto Github PK

Adaptation of SDV model (number of hidden layers for encoding and decoding, etc.) is done at the SDV model's construction, via arguments.
Models have different keyword arguments to specify their internal layout, therefore, every specific ASyH model needs to construct an argument list from the input data layout.
For keyword arguments, a dictionary should be used.

The dictionary can then be used as argument as in
sdv_model_constructor(**argument_dict).

Compare p-values of tests on input vs synthetic data.

Generating a report with ASyH 1.0.0 will fail with errors like:

Traceback (most recent call last):
  File "/beegfs/scratch/ASyH/ARX-ASyH-Comparison/ASyH-scripts/2-train+sample+report.py", line 35, in 
    report.generate(dataset, asyh_application.model.model_type)
  File "/beegfs/homes/hwilhelmi/.venvs/ASyH/lib/python3.9/site-packages/ASyH/report.py", line 49, in generate
    markdown = self.get_mark_down_report(dataset_name, sd_model_name, images)
  File "/beegfs/homes/hwilhelmi/.venvs/ASyH/lib/python3.9/site-packages/ASyH/report.py", line 107, in get_mark_down_report
    jinja_template = self._get_report_template()
  File "/beegfs/homes/hwilhelmi/.venvs/ASyH/lib/python3.9/site-packages/ASyH/report.py", line 124, in _get_report_template
    return env.get_template('report.j2')
  File "/beegfs/homes/hwilhelmi/.venvs/ASyH/lib/python3.9/site-packages/jinja2/environment.py", line 1010, in get_template
    return self._load_template(name, globals)
  File "/beegfs/homes/hwilhelmi/.venvs/ASyH/lib/python3.9/site-packages/jinja2/environment.py", line 969, in _load_template
    template = self.loader.load(self, name, self.make_globals(globals))
  File "/beegfs/homes/hwilhelmi/.venvs/ASyH/lib/python3.9/site-packages/jinja2/loaders.py", line 126, in load
    source, filename, uptodate = self.get_source(environment, name)
  File "/beegfs/homes/hwilhelmi/.venvs/ASyH/lib/python3.9/site-packages/jinja2/loaders.py", line 218, in get_source
    raise TemplateNotFound(template)
jinja2.exceptions.TemplateNotFound: report.j2

Calculate pairwise distance of each input data entry to each synthetic data entry, average.

Calculate the scalar products of each input data entry and each synthetic data entry, normalise, determine maximum

Pairwise correlations between all numerical variables for input and synthetic data.

Train on Real, Test on Real vs. Train on Synthetic, Test on Real.

The actual instantiation of the SDV model should be done just before we want to train with data, since we want to make sure we can adapt the argument list for the model constructor according to the input data and metadata.

The _train() method is defined in the generic Model class. The specific constructor should therefore be called with adapted arguments within the generic interface.

Thus, we should not provide a ready-made SDV model object to Model.init(), but instead provide a method calling the actual constructor with the specific arguments (as a dictionary, compare Issue #6).