v0.2.7

Added:
- Adding FWHM plugin
- Adding STD to FWHM
- Adding CVDiagnostics, to check correlations in classifier inputs and
  other variables.
- Can define new features from existing columns
- More PDFs in ModelFactory, HypExp, ModExp, etc
- to_yaml and from_yaml utilities to deal with pandas dataframes
- Adding hashing function, ot hash generic python objects.
- JSON loader added to existing JSON dumper

Fixed:
- CV predict testing module had `_get_models` renamed to `ut.get_models`

Improved:
- Requiring latest version of TF
- Improve fitting parameter ranges
- Can set log or linear y scale for fit plots
- Can preparametrize peaky PDFs with scale and resolution

0.2.6: Updating

- Improving parameter ranges for ModelFactory PDFs
- Better fitting messages
- Better fitting factory arguments, can control floating and shared
  parameters

0.2.5: Updates

- Bug: Plotter 2D was not calling base class initializer
- 2D plots can have z axis in log scale
- Can configure legend
- Can add statistics information in legend
- When plotting can skip vertical mass lines

- Sending outputs of tests to /tmp
- In ModelFactory mu and sigma parameter will always float, have _flt

- Configs in pkl file can be sent to YAML.
- Improve tests

0.2.4: Updates

- Adding version manager

0.2.3: Updates

Adding `add_column_with_numba` which is faster than old adder.

0.2.2: Improvements

- Better ROC curve
- Fix bug in annotations calculation
- Transforming numpy -> awk to go around awkward bug
- Moving to new awkward version
- Predictor will automatically define columns and clean nans based on
  config stored in model

0.2.1: New features

- Can save feature importance as latex
- Can save correlation matrix for training sample
- Adding Nan replacement latex table
- Preprocessing nans, replacing them
- Better score plots, show number of entries
- Fix Background rejection label in ROC curves
- Adding annotation to ROC curves

0.2.0: Improvements

- Documentation for GoFCalculator
- Sending outputs of tests to /tmp
- Testing 1D plot normalization
- Re-enabling normalization of histograms with Hist backend through
  weight normalization.
- Bug: CVPredictor was returning 2D array instead of 1D array of signal
  probabilities.
- Downgrading awkward due to bug: https://github.com/scikit-hep/awkward/issues/3373

0.1.9: New features

- Adding `minimizers` module with AnealingMinimizer
- Adding GofCalculator.
- Adding `ModelFactory` class

0.1.7: New features

- 1D plotter uses hist instead of matplotlib as backend
- Added ZFitPlotter