Improve Usability #62
Labels
Comments
|
This was referenced Apr 7, 2018
|
kimt33
added a commit
that referenced
this issue
Jun 22, 2018
1. Parameters are assumed to be given as a flattened array where the first matrix has the shape (4, 1, ndim), last matrix has the shape (4, ndim, 1) and the rest have the shape (4, ndim, ndim), where ndim is the selected dimension. 2. Methods, `get_occupation_indices`, `get_matrix_shape`, `get_matrix_indices`, `get_matrix`, and `decompose_index`, have been added to navigate the different tensors from the flattened parameter. Should a different parameter structure be used, only these methods should be changed. 3. `_olp` and `_olp_deriv` have been separated out of the `load_cache` in anticipation of usability improvements (Issue #62) Issues 1. `assign_params` and `load_cache` can be moved to the parent class
kimt33
added a commit
that referenced
this issue
Jul 1, 2018
What:
1. Move out _olp and _olp_deriv as a method of BaseWavefunction
(rather than functions hiding inside method _load_cache)
2. Remove creation of _cache_fns upon initialization
3. Remove creation of _cache_fns in load_cache
4. Prevent clearing cache when _cache_fns does not exist
5. Improve coverage of BaseWavefunction
6. Modify docstrings for clarity
Why:
1. Easier to document and to read. Unit testing will be easier. Can
remove repeated code.
2. Because not all wavefunctions will need to cache overlaps
3. Because consistency
4. Causes errors (especially since _cache_fns does not always exist)
kimt33
added a commit
that referenced
this issue
Jul 1, 2018
What: 1. Turn params_shape into an abstract property 2. Use params_shape instead of template_params to compute nparams in BaseWavefunction 3. Use nparams in BaseWavefunction.load_cache instead of params.size 4. Move around tests Why: 1. Every time param_shape is called, template_params needs to be constructed. For most wavefunctions, this is not a big problem, but some wavefunctions actually need to do a fair bit to obtain a template (e.g. APr2G). The consistency between params_shape and template_params will be checked implicitly through assign_params, which checks that the given parameters has the same shape as the params_shape. Fix up BaseWavefunction.nparams
kimt33
added a commit
that referenced
this issue
Jul 2, 2018
Can be used to generate arbitrary examples (within the scope of the
script).
What:
1. Add script for generating code that would run calculations.
Why:
1. Useful for making examples. Can be treated as an "input" to the
calculations.
kimt33
added a commit
that referenced
this issue
Jul 2, 2018
What: Add script for running calculations. Why: Since `wfns_make_script.py` already produces a script for running the calculations, the script is produced as a string then executed to run calculations. We could've hacked into the stdout to get the script as a string or saved to a temporary file, but it was easier to add an extra functionality to `wfns_make_script` to return a string which then gets executed.
kimt33
added a commit
that referenced
this issue
Aug 8, 2019
1. Parameters are assumed to be given as a flattened array of matrices of shape(nelec, nspin) 2. Denominators are differentiated from numerators by with attribute `numerator_mask` 3. Methods, `get_columns`, `get_matrix`, and `decompose_index`, have been added to navigate the different matrices from the flattened parametered. Should a different parameter structure be used (e.g. each matrix having different shape from (nelec, nspin)), then only these methods need to be changed. 4. `_olp` and `_olp_deriv` have been separated out of the load_cache in anticipation of usability improvements (Issue #62) Issues: 1. `load_cache` will need to move to the base parent class 2. `assign_params` is redundant (and can move to the parent class) 3. `np.logical_not` was used instead of the unary operator `~` or `-`
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Following changes are proposed for improving usability and ease of development
dtypeattribute in BaseWavefunction_olpand_olp_deriv) withinload_cacheof wavefunction classes to outside (for ease of testing) and wrap these functions withinload_cacheto removeselffrom parameters.template_params? add param shape and size as properties that depend ontemplate_params(and overwrite if desired)?OneRefApproxaround geminal instances rather than using multiple inheritancesThe text was updated successfully, but these errors were encountered: