-
Notifications
You must be signed in to change notification settings - Fork 4
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
100 refactor fitting for feature sets #101
base: master
Are you sure you want to change the base?
Commits on Jun 17, 2019
-
Initial framework for multiproperty fitting.
Begins solution for #100 with an initial solution. Also adds various supporting features to ProfileProperty and PropertyDict.
Configuration menu - View commit details
-
Copy full SHA for d168ca5 - Browse repository at this point
Copy the full SHA d168ca5View commit details
Commits on Jun 18, 2019
-
Testing for the interpolation and normalization.
Changes to the bayes testing in order to accomidate the change of the fitting to interpolate outside of the fitting process.
Configuration menu - View commit details
-
Copy full SHA for 1c9cd29 - Browse repository at this point
Copy the full SHA 1c9cd29View commit details -
Added tests for features used during fitting.
This includes a test for a new "feature_domain" property. This property is used to enable the testing of interpolated values prior to attempting to perform a fit.
Configuration menu - View commit details
-
Copy full SHA for 3f76322 - Browse repository at this point
Copy the full SHA 3f76322View commit details
Commits on Jun 19, 2019
-
Configuration menu - View commit details
-
Copy full SHA for 5222a06 - Browse repository at this point
Copy the full SHA 5222a06View commit details -
Update feature weighting to be based on errors.
This is done if errors are available. In either case weighting has a two norm of 1. Also updates modelling to not depend on properties having a length to use scalar properties.
Configuration menu - View commit details
-
Copy full SHA for 4c93c01 - Browse repository at this point
Copy the full SHA 4c93c01View commit details
Commits on Jun 20, 2019
-
Also allow PropertyDict.subset to take a single name since this is a potentially common occurance and if not handled correctly can provide strange results. Options were to throw an error or to correctly implement it.
Configuration menu - View commit details
-
Copy full SHA for ba3a10f - Browse repository at this point
Copy the full SHA ba3a10fView commit details -
Change fitting procedure to expose models.
This allows users to configure or adjust parameters prior to performing the fits. Adjusts the tests to match.
Configuration menu - View commit details
-
Copy full SHA for 9cd1db8 - Browse repository at this point
Copy the full SHA 9cd1db8View commit details -
Expose weighting option for fitting.
Also removed "unusable" guess function which would not work with the composite modelling used during fitting.
Configuration menu - View commit details
-
Copy full SHA for 69f836c - Browse repository at this point
Copy the full SHA 69f836cView commit details -
Configuration menu - View commit details
-
Copy full SHA for 18238b5 - Browse repository at this point
Copy the full SHA 18238b5View commit details
Commits on Jun 21, 2019
-
Described available parameters for the model.
Also fixed the probability when there is only one depth.
Configuration menu - View commit details
-
Copy full SHA for a539652 - Browse repository at this point
Copy the full SHA a539652View commit details -
Configuration menu - View commit details
-
Copy full SHA for 16cec72 - Browse repository at this point
Copy the full SHA 16cec72View commit details
Commits on Jun 24, 2019
-
Do not interpolate inplace by default.
Also addresses additional documenation changes for PR.
Configuration menu - View commit details
-
Copy full SHA for 9c2b2c5 - Browse repository at this point
Copy the full SHA 9c2b2c5View commit details -
Configuration menu - View commit details
-
Copy full SHA for 0b42551 - Browse repository at this point
Copy the full SHA 0b42551View commit details -
Use set_param_hint to be more lmfit idiomatic.
Also closes #102 by allowing global minimization method choice.
Configuration menu - View commit details
-
Copy full SHA for 25c4f5e - Browse repository at this point
Copy the full SHA 25c4f5eView commit details -
This closes #102 by providing tests for using it.
Configuration menu - View commit details
-
Copy full SHA for 322f1b2 - Browse repository at this point
Copy the full SHA 322f1b2View commit details -
Test TabulatedModel with more inputs.
Previously rewrote tests to exclude interpolation so undoing that change. Also, include testing the centering aspect of the model.
Configuration menu - View commit details
-
Copy full SHA for 4ab9ec2 - Browse repository at this point
Copy the full SHA 4ab9ec2View commit details
Commits on Jun 25, 2019
-
Configuration menu - View commit details
-
Copy full SHA for 6e353cc - Browse repository at this point
Copy the full SHA 6e353ccView commit details -
Begins code structure for fixing #103.
However no testing has been completed and the "legacy" multiproperty fitting has not been removed. Initial use with `fitting_both.py` script indicates successful fitting.
Configuration menu - View commit details
-
Copy full SHA for 78e0646 - Browse repository at this point
Copy the full SHA 78e0646View commit details
Commits on Jun 26, 2019
-
Fixes #103 by using new general modelling.
Replaces references to the MultiPropertyModel class for the new model creation. Adjustments to the test to reflect the change in model creation interface. Tests for acceptance of general models and tests for parameter naming are still to be completed.
Configuration menu - View commit details
-
Copy full SHA for fef9df2 - Browse repository at this point
Copy the full SHA fef9df2View commit details -
Configuration menu - View commit details
-
Copy full SHA for 92ff0b4 - Browse repository at this point
Copy the full SHA 92ff0b4View commit details -
Fix #103 by adding tests specific to new changes.
This goes beyond the integration tests already present to test specific expected behavior for the added general modelling.
Configuration menu - View commit details
-
Copy full SHA for c059277 - Browse repository at this point
Copy the full SHA c059277View commit details -
Update documentation regarding
models
.This is to reflect that functions should not be directly passed through to this parameter. It is expected that functions are first converted to lmfit.Model classes or instances.
Configuration menu - View commit details
-
Copy full SHA for 4a7c814 - Browse repository at this point
Copy the full SHA 4a7c814View commit details -
Fixes bug of model objects not copying.
When Model instances are passed instead of classes, the object was not properly handled which lead to each "use" of that model template to overwrite the previous uses.
Configuration menu - View commit details
-
Copy full SHA for 16301ed - Browse repository at this point
Copy the full SHA 16301edView commit details -
Configuration menu - View commit details
-
Copy full SHA for 05c693e - Browse repository at this point
Copy the full SHA 05c693eView commit details
Commits on Jun 27, 2019
-
Configuration menu - View commit details
-
Copy full SHA for 4d49da7 - Browse repository at this point
Copy the full SHA 4d49da7View commit details -
Change handling of variable argument type.
Attempt to be more "pythonic" by using try accept. Avoids using `isinstance` to check types. Generalizes accepted types to single object or iterables not just lists.
Configuration menu - View commit details
-
Copy full SHA for 45204c3 - Browse repository at this point
Copy the full SHA 45204c3View commit details
Commits on Jun 28, 2019
-
Configuration menu - View commit details
-
Copy full SHA for 0b3cf9b - Browse repository at this point
Copy the full SHA 0b3cf9bView commit details
Commits on Jul 1, 2019
-
Fix docs. Change
subset
None
behavior.PropertyDict no longer expects `None` for names to resturn self. Instead, expect user to pass self.keys() for a complete subset.
Configuration menu - View commit details
-
Copy full SHA for e1aa603 - Browse repository at this point
Copy the full SHA e1aa603View commit details
Commits on Jul 3, 2019
-
Adds optional filtering when taking a subset of PropertyDict. Adds periods to documentation. Change calls of `super` to use no arguments. Attempt to optimize property interpolation by only creating the interpolator once instead of on each call.
Configuration menu - View commit details
-
Copy full SHA for 172feb3 - Browse repository at this point
Copy the full SHA 172feb3View commit details -
Configuration menu - View commit details
-
Copy full SHA for 909ab07 - Browse repository at this point
Copy the full SHA 909ab07View commit details -
Configuration menu - View commit details
-
Copy full SHA for 495992c - Browse repository at this point
Copy the full SHA 495992cView commit details
Commits on Jul 5, 2019
-
Use random seed to reduce test flaky-ness.
This leads to reliability that failed tests are not due to just randomness. This raises the issue that the occasional failures from the randomness should be classified, fixed, and have dedicated tests. Hypothesis tests may be useful for this.
Configuration menu - View commit details
-
Copy full SHA for 0a66bf8 - Browse repository at this point
Copy the full SHA 0a66bf8View commit details -
Persist interpolator kws through modifications.
Allow user settings for interpolators to persist through property data modification by changing how interpolators are created. These settings are now saved into an attribute of the property.
Configuration menu - View commit details
-
Copy full SHA for f7817f0 - Browse repository at this point
Copy the full SHA f7817f0View commit details -
Patched models to use modified _residual.
This is to enforce structure probabilities sum to 1. Also rewrote global minimization testing to actually make sense.
Configuration menu - View commit details
-
Copy full SHA for 08b93e4 - Browse repository at this point
Copy the full SHA 08b93e4View commit details
Commits on Jul 8, 2019
-
Change modelling to no longer bound probs.
Restructure modelling to focus on linear "mx + b" models. Change to using proportions that are converted to probabilies for interpretation. Remove tabulated only modeling and instead allow choice of tabulated modelling in the restructured modelling options.
Configuration menu - View commit details
-
Copy full SHA for 4d0e6c1 - Browse repository at this point
Copy the full SHA 4d0e6c1View commit details -
Configuration menu - View commit details
-
Copy full SHA for aaefb41 - Browse repository at this point
Copy the full SHA aaefb41View commit details -
Configuration menu - View commit details
-
Copy full SHA for d84858a - Browse repository at this point
Copy the full SHA d84858aView commit details -
Configuration menu - View commit details
-
Copy full SHA for 029a686 - Browse repository at this point
Copy the full SHA 029a686View commit details
Commits on Jul 9, 2019
-
Update code and documentation of model creation.
Updates were to improve clarity.
Configuration menu - View commit details
-
Copy full SHA for 6797190 - Browse repository at this point
Copy the full SHA 6797190View commit details -
Configuration menu - View commit details
-
Copy full SHA for 748a148 - Browse repository at this point
Copy the full SHA 748a148View commit details
Commits on Jul 10, 2019
-
Address overwriting parameter hints.
This involves being more verbose and adding tests for these parameter hints. To simplify this, this changes the tabulated model to use the same parameter names to simplify the logic to hard coded parameter names.
Configuration menu - View commit details
-
Copy full SHA for d254652 - Browse repository at this point
Copy the full SHA d254652View commit details
Commits on Jul 12, 2019
-
Fix bug where tabulated choice was ignored.
The `use_tabulated` was not being passed down from create to depth to the building block functions.
Configuration menu - View commit details
-
Copy full SHA for be0c8af - Browse repository at this point
Copy the full SHA be0c8afView commit details
Commits on Jul 29, 2019
-
Distinguish between composite and simple tabulate.
The difference places restrictions on the composite to not allow simultaneous centering and interpolation to different qvalues. The restriction is in place because from the experimental feature vector it is difficult to determine what section should be used as the new qvalues. Therefore require user to interpolate before and simply allow centering in composite.
Configuration menu - View commit details
-
Copy full SHA for 855ed21 - Browse repository at this point
Copy the full SHA 855ed21View commit details