- Fixed an issue with the intra-annual metrics adapter when an empty time series is passed (PR #307)
- Metapackage updated (pyscaffold 4.5) (PR #307)
- C modules were outdated and could not be compiled, therefore rebuilt (PR #307)
- Adapters for metric calculatores were introduced. The here implemented adapters compute metrics based on temporal subsets of the time series, which can be used for rolling window metrics, seasonal metrics or multiple arbitrary sub-periods. (PR #266)
- Fixed a bug where the status code of a successful Triple Collocation run was still set to -1 ("unknown error"). (PR #301)
- Add kwarg to time_series.anomaly.calc_climatology to compute the standard deviation of the climatology (e.g. for anomaly z-scores) (PR #290)
- Adding scikit-learn as installation dependency. (PR #298)
- fixing setting of status code for bootstrapped triple collocation (PR #285)
- Exclude netCDF4 at version 1.6.2 due to bug when writing strings with zlib compression (PR #288)
- added test for data_manager_failed (PR #287)
- read_ts is not longer used in pytesmo (PR #289)
- support for max_dist keyword in validation framework to limit comparisons to close data points (PR #291)
- Reimplementation of CDF matching (PR #259). This removed all other implementations, i.e.
lin_cdf_matchandcdf_beta_match. These now are either removed (lin_cdf_match), or point to the new implementation (cdf_beta_match)
- option to ignore NaN values in filter fields (PR #282)
- fixing rounding error issues in metric calculations (PR #281)
- fixed bugs in resample_mean (PR #262)
- Integration tests with new ISMN Interface (PR #263)
- Add option to Validation Framework to ignore errors in
Validation.calc()(PR #263) - ipynb files from docs/examples are now also used as (optional) tests (PR #263)
yapffor code formatting (see developers guide) (Fix #248, PR #263)- validation framework option to force dataset combinations that include reference dataset updated (PR #265)
- Added TimestampAdapter to the validation framework to deal with datasets that have different date/time specification fields (PR #268)
- use oldest-supported-numpy in pyproject.toml to avoid errors due to incompatible numpy versions used in build-time.
- rollback to old C-files due to numpy compatibility issues
- monthly climatologies
wraparound=Trueis the new default in the climatology calculation- better error handling in validation framework
- Merge pull request #249 from pstradio/monthly_clim
- Monthly climatology
- monthly climatologies
- wraparound=True is the new default in the climatology calculation
- better error handling in validation framework
- Add option for custom reading method name to all reading adapters (PR #245)
- MonthMetricsAdapter for new PairwiseIntercomparisonMetrics and TripleCollocationMetrics calculators (PR #246)
- fixed some bugs in the confidence interval calculation, for some metrics only bootstrapped CIs are available now
- fixed bug in rolling metric calculation (now also works with float32 input data)
- minor bug fixes
- removed
ascatandismnfrom the package dependencies - fixed a few bugs in
pytesmo.temporal_collocation.combined_temporal_collocation. - updated documentation
- added Python 3.9 to CI
Analytical and bootstrapping confidence intervals for metrics (PR #206). This includes some changes to the existing implementations (all old implementations are still available, but deprecated)
all pairwise metric functions take two arrays as input and return a single value
the correlation metrics (
pearsonr,spearmanr,kendalltau) have new versionspearson_r,spearman_r, andkendall_tauwhich only return the correlation value, but not the p-value. The old functions have been deprecated. For calculating correlation + p-value, it is advised to usescipy.stats.pearsonr,scipy.stats.spearmanr, andscipy.stats.kendalltau. Instead of p-values, confidence intervals for the correlation coefficients could be obtained with:r, lower, upper = with_analytical_ci(pearson_r, x, y)
pytesmo.metrics.tcol_errorandpytesmo.metrics.tcol_snrhave been deprecated. Usepytesmo.metrics.tcol_metricsinstead (which is simply a renaming oftcol_snr).pytesmo.metrics.msehas been deprecated. There is a new, much faster implementation available (pytesmo.metrics.mse_decomposition). Individual values of the components can be calculated withpytesmo.metrics.mse,pytesmo.metrics.mse_corr,pytesmo.metrics.mse_bias,pytesmo.metrics.mse_var.
Removed dependency on deprecated Numpy API
added mean resampling in temporal collocation
updated to
ascatversion 2.0
- Drop Travis CI and Appveyor CI, replaced by Github Actions ci.yml workflow
- New implementation of temporal collocation based on pd.reindex (PR #204): :py:func:`pytesmo.temporal_matching.temporal_collocation`
- CDF matching function updated
- min_obs keyword added to MetricsCalculators
- Include more combinations in validation framework, raise error if n < n_datasets
- n_wise_apply now can handle (drop) nans in a passed data frame correctly.
- TC metrics calculator has now option to calculate metrics for reference
- Fix deprecation warnings in anomaly adapter (Issue #198)
- Change combinations created by val framework, catch cases where scaling ref not in combinations (Issue #187)
- Update pyscaffold v3.2.3
- Rolling Prearson's R and RMSD metrics calculator
- New results manager to handle rolling metrics
- Add deprecation warning for colormaps module
- Move timedate module to cadati, import from there
- Filter warnings in places where they are safe to ignore and not caused by pytesmo
- Update Docs and ipython notebooks
- Update Triple Collocation metrics and metrics caluclator
- Update metrics apply function, use dictionaries instead of dataframe for results
- Remove io module and grids.grid.py and grids.netcdf.py modules
- Move resampling module to repurpose package (https://github.com/TUW-GEO/repurpose)
- Documentation updates
- Change naming convention for Intercomparison metrics calculator results
- Use kdtree from pykdtree instead of scipy (faster)
- Add MetadataMetrics to metrics calculator
- Remove pybufr-ecmwf, use generic test data for testing spatial resampling
- Add option for filling no data values to moving average
- Add option for minimum observations to moving average
- Default behavior of climatology calculation will now fill no_data values Nothing is changing if there are no no_data values in the time series
- Add more metrics calculators
- Remove pytesmo.io.ismn module and move it to the ismn package
- Update tests for ascat v1.0 package
- Rename arguments for metrics functions to be more general
- Split linreg scaling function to determine and apply corrections separately
- Compatible with Python 3.7 on Windows
- Update readme
- Fix bug in exponential filter when first value is a NaN value
- Add extendent collocation metric
- Fix initial value for exponential filter
- Fix #123
- Adapt validation framework examples to new ASCAT package version.
- Adapt ERS reader to new ASCAT package version.
- Make validation framework work with datasets that contain NaN columns.
- Make validation framework work with pygeobase.object_base.TS objects and subclasses.
- Add scaler classes to the validation framework making it possible to use e.g. stored CDF parameters during validation.
- ensure_iterable does no longer take a single string as an iterable. We want to keep the string as one object.
- Add respect leap years option for climatology calculation.
- Compatible with Python 3.6
- Add additonal functions for working with dekads.
- Refactor resampling routine to be more modular and better usable outside of pytesmo.
- temporal matching performance improvement of approx. 50%
- Add functions for handling dekadal dates. See
pytesmo.timedate.dekad.
- Fix metadata for new version of pypi.
- Add return_clim keyword to anomaly calculation. Useful for getting both anomaly and climatology in one pandas.DataFrame. Also used in time series anomaly plot.
- Fix bug in julian2date which led to negative microseconds in some edge cases.
- Moved the ASCAT readers to the ascat package. The functionality is
the same, just replace
import pytesmo.io.sat.ascatbyimport ascatand everything should work the same as before. - The H07 reader now returns also ssm mean as a value between 0 and 100. Before it was between 0 and 1 and inconsistent with the other ssm values.
- Fix small bug in julian date calculation and add tests for it.
- Add hamming window to resample module
- Fix bugs when the validation framework encountered empty datasets for various reasons.
- Add dataset adapters for masking and anomaly calculation.
- Improve performance of moving average calculation and ISMN readers.
- Fix bug in jobs argument passing to Validation class.
- Add support to use a pre initialized DataManager instance in the Validation class.
- Add support for per dataset reading method names in the DataManager.
This relaxes the assumption that every dataset has a
read_tsmethod.
- Fix bug in temporal resampling if input was a pandas.Series
- Major refactoring of validation framework. Please see updated documentation and example for detailed changes. The most important breaking changes are:
- 'type' is no longer used in the dataset dictionary.
- the temporal matcher does no longer need to be specified since a reasonable default was developed that should handle most cases
- metrics calculators are now given as dictionaries of functions. Please see the docs for an explanation and an example.
- cell_based_jobs keyword was removed in favor of a more general definition of jobs.
New features are the possibility to use unrelated masking datasets and the possibility to temporally match any number of datasets and give them in sets of k datasets to multiple metric calculators.
- Changes in the scaling module, escpecially CDF matching. The new CDF scaling module is more modular and does not make any assumptions about how unique the percentiles for the CDF matching have to be. CDF matching now returns NaN values if non unique percentiles are in the data. There are new functions that rescale based on pre-calculated percentiles so these can be used if the user wants to make sure that the percentiles are unique before matching.
- Fix bug in validation framework due to error prone string formatting in warnings.
- Remove grid functionality. Use pygeogrids from now on.
- Fix bug in moving average calculation when input had size 1.
- Add recursive calculation of Pearson correlation coefficent.
- Change H-SAF reading interface to use pygeobase consistently. This
changes the interface slightly as the
read_imgmethod is now called justread - H07 reader now returns more variables.
- Resampling interface now respects dtype of input data.
- Improvements in ISMN plotting interface make it possible to use the plot not only show it.
- make sure that climatologies are always 366 elements
- add new options to climatology calculation for filling NaN values
- add option to climatology calculation for wraparound before the smoothing
- fix bug in anomaly calculation that occurred when the climatology series had a name already
- add option in anomaly calculation to respect leap years during matching
- improve testing of scaling functions
- add linear CDF scaling based on stored percentiles
- add utility function for MATLAB like percentile calculation
- add utility function for making sure elements in an array are unique by using iterative interpolation
- fix #63 by moving data preparation before period checks
- fix bug in exponential and boxcar filter. Problem was that nan values were not ignored correctly
- add option to temporal resampling to exclude window boundaries
- fix #48 by reintroducting netcdf imports
- fix #60 by importing correctly from pygeogrids
- fix #56 by allowing read_bulk keyword for ASCAT_SSM
- fix #58 by using cKDTree keyword if available
- lookup table indexing fixed, see #59
- hotfix for temporal resampling problem when time series where of unequal lenghts
- added validation framework and example on how to use it
- fix bug (issue #51) in temporal matching
- added test data as git submodule
- added calculation of pearson R confidence intervals based on fisher z transform
- ISMN reader can now get the data coverage for stations and networks
- ISMN interface can now be restricted to a list of networks
- added python3 support
- moved grid functionality to pygeogrids package, pytesmo grids are deprecated and will be removed in future releases
- include triple collocation example and improve documentation see issue #24
- fixed ASCAT verion detection for latest H25 dataset WARP55R22
- added example for Soil Water Index calculation
- moved to pyscaffold structure
- added tests for modules
- added grid generation routines
- fix for issue #15
- updated classes to work with new base classes, does not change API
- added travis CI support
- changed theme of documentation, and enabled read the docs
- added grouping module
- fixed bug that lead to old grids without shape information not loading
- added functionality to save grid as 2 dimensional array in grid.netcdf if grid is regular and shape information is given
- added readers, tests and examples for H-SAF image products H07, H08 and H14
- added resample method that makes using pyresample a easier for the dictionary structure that pytesmo uses for image data
- added colormap reader for custom colormaps
- fixed bug in grid.nearest_neighbour that caused different results on different systems. Radians are now always calculated at 64bit accuracy
- ISMN routines now read the new ISMN download format
- df_metrics.bias now also returns a namedtuple
- Reader for different versions of netCDF H25 HSAF product
- added functionality to save grid definitions to netCDF files
- Fixed Bug that masked all data if snow probabilities did not exist
- Added tests
- Added readers for netCDF H25 HSAF product
- Added readers for netCDF ERS soil moisture product
- Added general grid classes
- Performance improvements for anomaly and climatology calculation through usage of cython
- Introduced df_metrics module for convienent calculation of metrics for data saved in pandas.DataFrames