Feature/refactor impact calc (#436)

* Add methods to set imp vars from imp mat

* Cosmetics

* Change set to calc

* Ordering the code

* Make init

* Add haz_type as property

* Add centroid columns name in exposures as property

* Add overwrite variable to assign_centroids

* Add method affect_values_gdf

* Remove warning impf at int=0

* First commit - overhaul impact

* Add insurance layers

* Add methods get mdr, paa, fraction

* Change sum to np.sum

* Add affect_total_value

* Reduce gdf to minimum necessary

* Correct exposure gdf column selection

* Update docstrings

* Fix bug at_event / eai_exp

* Refactoring of methods to simplify code

* Use generator instead of list

* Impact: improve performance by avoiding pandas operations

* Add notimplementederror for mdf(0)

* Toggle logger warning for speed test

* Fetch only assigned impf

* Return the reduced exp indices only

* Remove 0 from insured impacts

* Stich imp mat from generator

* Rename impat_mat_list to _gen

* Avoid generator to list conversion in insured risk

* Rename exp_iimp

* Add impact_calc class

* Correct function signature

* Adjust class name

* Correct indent

* Typo: stich -> stitch

* PEP8

* impact: pydoc cosmetics

* Include all helper methods in ImpactCalc

* Remove unused module

* Improve cosmetics

* Improve docstring

* Import new class in init

* Compute insured impact from existing imp_mat if available

* Make two test files

* Add _return_impact method

* Return impact object

* Make risk_metrics a class method

* Separate Impact and ImpactCalc

* Add deprecation warnings

* Return min exp gdf instead of df

* Improve cover and deductible

* Rename imp_mat to mat

* Add tests for most exposed methods

* Remove calc tests

* Add test overwrite centroids flag

* Remove warning for impf nonzero at zero

* Add docstrings

* Update test precision to account for different floating number

* Fix bug in insure mat

* Make 'imp_mat' an empty matrix by default

* Use empty csr matrix instead of 'None' as default argument for
  'imp_mat' in 'Impact' and 'ImpactCalc'.
* Consistently check for matrix contents via its 'size' attribute.

* Avoid mutables in Impact and ImpactCalc signatures

* Add test case for ImpactCalc.impact_matrix

Use unittest.mock to mock behavior of hazard object.

* Add input attribute consistency checks to init

* Add unfinished impact init test

* Set default imp_mat to None

* Update init consistency tests

* Add from_eih test

* Add test risk transfer and residual risk

* Add test impact_per_year

* Cosmetics

* Remove unsued method

* Add test for affected value

* Add property tests

* Add test get_mdr

* Add test get paa

* Add test get fraction

* Add docstring

* Add chunk method

* Update name cent to centr

* Update docstring

* Use np.array_split for chunks

* Update naming

* Simplify argument chunker

* Add test case for ImpactCalc.imp_mat_gen

* Add test for ImpactCalc.stitch_impact_matrix

* Add test for ImpactCalc.apply_deductible_to_mat

* Fix bug and formatting in ImpactCalc.imp_mat_gen

* Use the original impf_id instead of relying on the ID of the returned
  impact function object.
* Remove superfluous whitespace

* Add test with fraction not equal 1

* Improve error message in impact matrix chunking

* test_impact_calc: persistently revert the CONFIG changes in tearDown

* Replace instances in CONFIG instead of mocking

Explicitly replace instances in the CONFIG tree with new/old values
in the TestImpactMatrixGenerator test case, because CONFIG cannot be
deepcopied. This avoids errors in tests that are executed after this
test case.

* Add test for ImpactCalc.stitch_risk_metrics and update docstring

* Add deepcopy import to test_impact_calc

* Add test for ImpactCalc.insured_mat_gen and update docstring

* Add test for ImpactCalc.impact_matrix

* Add TestInsuredImpactMatrixGenerator to test loader

The test was previously not added to the loader, but this only had an
effect when the file was executed directly (as __main__).

* Add test for ImpactCalc._return_matrix

* Update docstring

* pydoc cosmetics

* pydoc cosmetics

* pydoc consolidation

* Improve attribute error message from_eih

* Update from_eih tests

* impact_calc.ImpactCalc: remove imp_mat attribute
turn n_exp_pnt and n_events into properties
set cover and deductible to None if not set (instead of an empty array)

* ImpactCalc.impact: re-establish save_mat argument

* Set assigned_centroids overwrite=False

* (Re)-Set ignore zero to False

* Update deprecation warning message

* Correct bug wrong exp_idx

The exposures index from the impact matrix generator was indexed for the reduced exposures only, instead of the total exposures. This leads to wrong indexing.

* Make new index in case the index is not ordered

* Slice numpy instead of gdf

* Add _orig_exp_idx to tests

* Add tests with Flood

* Return empty impact if no exposures matching hazard

* Repalce squeeze by ravel for atevent single event

* Add _orig_exp_idx as argument from beginning on

* Update comment cosmetics

* Reduce lines to max length

* Replace generator by list due to deprecation in np.hstack

* test_impact_calc: organize test data

Co-authored-by: Chahan Kropf <chahan.kropf@posteo.com>
Co-authored-by: Thomas Vogt <thomas.vogt@pik-potsdam.de>
Co-authored-by: emanuel-schmid <schmide@ethz.ch>
Co-authored-by: Lukas Riedel <34276446+peanutfun@users.noreply.github.com>

climada/engine/__init__.py

@@ -20,3 +20,4 @@
 """
 from .impact import *
 from .cost_benefit import *
+from .impact_calc import *

climada/engine/forecast.py

@@ -1214,7 +1214,7 @@ def plot_hexbin_ei_exposure(self, run_datetime=None, figsize=(9, 13)):
             The default is (9, 13)
         Returns
         -------
-        axes : cartopy.mpl.geoaxes.GeoAxesSubplot
+        cartopy.mpl.geoaxes.GeoAxesSubplot
         """
         # select hazard with run_datetime
         if run_datetime is None: