Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

0.15.1 - 2026-05-12

Fixed

  • 404 Error for examples launch binder badge

0.15.0 - 2026-05-12

Added

  • New captest.captest module. Exports CapTest, TEST_SETUPS, the three scatter callables scatter_default, scatter_etotal, scatter_bifi_power_tc, and the helpers perc_wrap, print_results, highlight_pvals, load_config, resolve_test_setup, validate_test_setup.

  • New CapTest class (captest.captest.CapTest) — Binds a measured and a modeled CapData together, holds every test-level setting (regression formula, reporting-conditions recipe, filter bounds, nameplate, tolerance, calc-params scalars), and drives the full capacity-test workflow from a single object.

  • New TEST_SETUPS registry of named regression presets shipped with pvcaptest. Four presets ship: e2848_default (default ASTM E2848), bifi_e2848_etotal (bifacial ASTM with e_total), bifi_power_tc (temperature-corrected bifacial power ~ poa + rpoa), and e2848_spec_corrected_poa (ASTM with a spectral-correction factor applied to POA irradiance).

  • New process_regression_columns method on CapData, which recursively walks the regression_cols dictionary.

  • New calcparams module with simple functions that return calculated regressors. E.g., the e_total function calculates total POA irradiance from front and rear POA.

  • New captest.calcparams functions for spectrally corrected irradiance tests: apparent_zenith, apparent_zenith_pvsyst (handles the PVsyst half-hour timestamp shift internally), absolute_airmass, precipitable_water_gueymard, scale, spectral_factor_firstsolar, multiply, and the named alias poa_spec_corrected.

  • New captest.calcparams functions to support the default TEST_SETUPS: e_total, rpoa_pvsyst, avg_typ_cell_temp, cell_temp, bom_temp, power_temp_correct.

  • CapTest.from_yaml / CapTest.to_yaml for a curated round-trip through a yaml config file.

  • New ScatterPlot and ScatterBifiPowerTc classes in captest.plotting that are used with the shipped scatter callables (scatter_default, scatter_etotal, scatter_bifi_power_tc).

  • POA vs power scatter plots can be split so morning and afternoon points are a different color or different marker (glyph).

  • POA vs power scatter plot function includes kwargs to switch between plotting raw power or temperature corrected power.

  • Documentation: new user-guide page user_guide/captest describing use of CapTest class.

Changed

  • Breaking: The agg_sensors method no longer updates the regression_cols dictionary. This functionality has been moved to process_regression_columns.

  • CapData.filter_time now defaults end to the last timestamp of data_filtered when only start is provided without days, and defaults start to the first timestamp of data_filtered when only end is provided without days.

  • Breaking: the module-level cross-CapData functions capdata.captest_results, capdata.captest_results_check_pvalues, capdata.pick_attr, capdata.get_summary(*args), capdata.overlay_scatters, capdata.determine_pass_or_fail, and plotting.residual_plot have been removed. Equivalent methods now live on CapTest (ct.captest_results(), ct.captest_results_check_pvalues(), ct.get_summary(), ct.overlay_scatters(), ct.determine_pass_or_fail(), ct.residual_plot()). pick_attr is superseded by the rep_cond_source parameter on CapTest. Single-CapData cd.get_summary() is unchanged.

  • Breaking: The CapData.rep_cond method has had the calculation of reporting conditions for multiple time periods moved to a separate method - rep_cond_freq.

  • Breaking: CapData.rep_cond and CapData.rep_cond_freq are now formula-agnostic: they derive rhs variables from regression_formula via util.parse_regression_formula.

  • CapData.scatter and CapData.scatter_hv are now formula-agnostic thin wrappers: they resolve the y (lhs) and x (first rhs) columns from regression_formula rather than hardcoding "power" and "poa". For non-default regression presets prefer CapTest.scatter_plots, which picks the correct scatter callable (single or multi-panel) from the resolved preset.

  • In filter_irr the string to use the reporting irradiance as the reference irradiance to filter around has been changed from rep_irr.

Convention

  • The left-hand-side key of the regression formula is always "power" across shipped TEST_SETUPS presets, even when the formula regresses a derived quantity like temperature-corrected power. Code that hardcodes "power" as the lhs key continues to work with all shipped presets.

0.14.0 - 2026-04-07

Added

  • perf_ratio_temp_corr_nrel accepts measured back-of-module temperature via temp_bom parameter, skipping the BOM calculation from ambient temp and wind.

  • perf_ratio_temp_corr_nrel accepts single_irr_weighted_temp parameter to use a single irradiance-weighted cell temperature for the correction.

  • Plotting function residual_plot to create plots of residuals vs regression parameters.

  • load_data can be used with the default file_reader to load csv files from an S3 bucket. Requires the new optional dependency s3fs.

  • agg_sensors method verbose option, which prints summary of aggregations.

  • Column groups are now accessible as attributes of CapData instances (e.g., cd.poa returns the data for the poa group). Attributes are created automatically when load_data is called or when CapData.create_column_group_attributes() is called manually.

  • CapData.rename_cols() method to rename columns consistently across data, data_filtered, and column_groups.

  • agg_sensors supports nested subgroup aggregation: pass a dictionary as the value for a group in agg_map to first aggregate subgroups and then aggregate the subgroup results. Aggregated columns are added to a new "agg" key in column_groups and are accessible as CapData attributes prefixed with aggs_.

  • CapData.plot() and plotting.plot() accept a plot_defaults_path parameter to explicitly specify the path to the plot defaults JSON file.

Changed

  • DataLoader.load now has a summary parameter (default True) that controls file loading status output. verbose default changed from True to False and now enables both summary output and detailed reindexing information.

  • DataLoader._reindex_loaded_files and DataLoader._join_files renamed to reindex_loaded_files and join_files (public API).

  • load_data now skips sort, drop_duplicates, and reindex when no data is loaded and issues a warning instead of raising an error.

  • Failed-to-load troubleshooting message now includes kwargs passed to file_reader.

  • filter_clearsky no longer accepts window_length as a direct parameter. It now defaults to infer_limits=True (pvlib recommended, Reno 2016). Pass window_length and other pvlib detect_clearsky parameters via **kwargs.

  • filter_missing refactored to use the floc indexer; the columns parameter now accepts column group keys in addition to column names.

Fixed

  • Removed stale references to the removed irr_rc_balanced function from docstrings of perc_bounds, rep_cond, and predict_capacities. These now reference ReportingIrradiance, the class that replaced it.

  • rep_cond docstring now documents that when irr_bal=True, the reporting temperature and wind speed are the means of the data within the balanced irradiance band.

  • CapData.plot() now reads and writes a per-instance defaults file named plot_defaults_{name}.json instead of the shared plot_defaults.json, preventing conflicts when multiple CapData objects are used in the same session (issue #117).

  • Columns listed in the plot defaults file that are no longer present in the data are ignored with a UserWarning rather than causing an error. If all stored columns are missing, the dashboard falls back to auto-detected default groups (issue #115).

  • Fixed inaccurate docstring note: the plot defaults file is resolved relative to the current working directory, not the calling file’s directory.

  • filter_clearsky positional arg bug where window_length was passed as infer_limits to pvlib’s detect_clearsky.

  • filter_outliers now detects NaN values in the poa and power columns before fitting the EllipticEnvelope. A UserWarning is issued and filter_missing is automatically called on those columns; both are recorded in the filter summary.

  • filter_power no longer incorrectly removes rows with NaN values when filtering on a multi-column power group.

  • DataLoader.load no longer raises UnboundLocalError when all files fail to load; sets data to None and issues a warning.

  • reindex_datetime now handles duplicate indices by dropping duplicates and warning instead of raising a ValueError.

  • filter_irr no longer uses pandas query(), fixing a compatibility error with pandas v2.2.3.

0.13.4 - 2026-02-17

Changed

  • Removed upper version limits for numpy, pandas, and param dependencies

  • Replaced use of deprecated copy kwarg removed in numpy 2.4

  • Removed use of pytz package and refactored timezone handling

Fixed

  • p-value checking for pandas 2 and 3 compatibility

  • param 2.0 deprecation for passing positional arg to ClassSelector

  • Various deprecation warnings in tests

  • Errors due to pandas 3.0 API changes, compatability for copy-on-write changes

0.13.3 - 2026-02-06

Changed

  • Migrated from versioneer to setuptools-scm for git-based version management

  • Migrated from setup.py to modern pyproject.toml with [project] section

  • Added uv for dependency management and development workflows

  • Updated version access to use importlib.metadata.version()

  • Updated numpy version constraint to <2.4.0 for compatibility, temporarily

  • Updated param version constraint to <2.0 for compatibility

  • Updated pandas version constraint to <3.0 for compatibility, temporarily

  • Updated development status classifier to “4 - Beta”

  • Simplified license metadata format in pyproject.toml

  • Moved testing and documentation dependencies to dependency-groups

  • Updated read the docs configuration to use uv

Removed

  • setup.py (all configuration moved to pyproject.toml)

  • Versioneer dependency and configuration

  • scipy dependency (replaced scipy.stats.mode with pandas.Series.mode in July 2023, unused imports removed January 2026)

Added

  • setuptools-scm for automatic version generation from git tags

  • Modern pyproject.toml with complete project metadata

  • Just command runner (justfile) with development recipes

  • Just recipes for testing package builds and version checking

  • Just recipe for building docs

  • MAINTAINER.md documentation for project maintainers

0.13.2 - 2025-07-17

Fixed

  • Change to nan for compatability with numpy 2.0

  • Update file_reader to warn but load empty csv files

0.13.1 - 2024-06-05

Fixed

  • Issue 112 - plotting tool would plot only last curve for plots with more than 25 curves.

0.13.0 - 2024-05-05

Added

  • load_data can read site location and system information for generating clear sky irradiance from a json or yaml file.

  • New plotting module with a plot function which replaces the CapData.plot method. The new plot function creates a panel dashboard with expanded functionality. Internally, removes the plotting dependency on the CapData.trans_keys attribute.

  • loc and floc can be used to retrieve the regression columns, similar to previous rview functionality by using regcols.

Changed

  • Removed the CapData trans_keys attribute, which was a copy of the column_groups keys and would be modified by the old CapData.plot method. trans_keys made it difficult to create a CapData object without using the load_data function.

  • Replaces all uses of view and rview with loc and floc.

  • Updated loc and floc to always return a DataFrame. Previously these would sometimes return a Series.

  • load_data now checks if any individual files were loaded when loading multiple files from a directory.

  • Adds underlay curve of unfiltered power to the linked timeseries created when calling scatter_hv with timeseries=True.

  • Changes selected points on scatter plot and linked timeseries produced by scatter_hv to red.

  • Tolerances may now be fractions eg ‘- 3.5’

  • The plotting methods scatter_hv, scatter_filters, and timeseries_filters do not require a column labeled ‘index’ with string datetimes in the data DataFrame anymore. Also, the index of the data DataFrame does not need to be named ‘Timeseries’.

  • Removes the add_index_col kwarg option from util.reindex.

0.12.0 - 2023-08-27

Added

  • Added a user guide section to the documentation with an overview and bifacial tests section.

  • Added verbose kwarg to the DataLoader load method, which prints which files have been loaded

  • DataLoader load method has new kwarg print_errors

  • DataLoader load method stores list of paths of files that fail to load in failed_to_load attribute

  • CapData method to save column_groups dictionary to excel

  • Github action to build and publish to PyPI on tags like v* pushed from master

Changed

  • Updates to make pvcaptest compatible with pvlib 0.10 and scipy 1.11

  • Update to make pvcaptest compatible with bokeh v3.0.0, change plot_width and plot_height to width and height

  • Make bokeh v3 minimum version

  • Drop support for python 3.7

  • Drop to_numeric from io.file_reader

  • CapData.column_groups is instance of ColumnGroups now when group_columns is a callable

  • Changed defaults for the CapData plot method to single column

0.11.2 - 2023-04-20

Added

  • Adds CI testing across python 3.7 to 3.11 on OSX, Linux, and Windows using Github actions.

Changed

  • Re-organized project root directory to place captest package under a src directory, see PR #83.

  • Re-organized the extras for pip installation into optional, test, docs, and all. Updated the installation section of the docs to reflect change.

  • Updated installation instructions with links to pandas and conda docs for more information on environments.

  • Re-organized installation instructions.

  • Changed instructions on creating a conda env for pvcaptest to directly download the env yaml file from the repository rather than the whole repository.

  • Clean up RTD configuration, particularly added project directory to PYTHONPATH in the docs build environment, so the docs build against checked out version instead of installed version from conda.

  • Updates to documentation to add new modules, remove history section of releases, and remove references to Travis CI.

Fixed

  • Cleaned up issues in tests found after re-implementing CI on other platforms.

0.11.1 - 2023-04-09

Added

  • Added new dependencies - colorcet, param

  • Added openpyxl as an optional dependency

  • Loads bokeh as holoviews extension in captest imports so user doesn’t need to

Changed

  • Removed use of hvplot, replaced with holoviews to avoid adding unnecessary dependency. PR #73.

Fixed

  • Typo in the import check of panel

0.11.0 - 2023-04-07 [YANKED]

Added

  • Added columngroups module with a ColumnGroups class that extends python dictionaries to include each column group as an attribute and repr is formatted for easy reading.

  • Created the loc and floc callables for CapData, which allow easier access to columns of the data and data_filtered attributes, repsectively. Will replace view and rview method.

  • Added read_json function to util module.

  • Added read_yaml function to util module.

  • Added the io module with functions to load data and return instances of CapData containing the loaded data.

  • Added the ReportingIrradiance class for calculations of a reporting irradiance that is between the 40th and 60th percentile. ReportingIrradiance.plot includes a dashboard type view of the selected and possible reporting irradiance values and method to save possible reporting conditions table to csv.

  • Added the spatial_uncert and expanded_uncert methods. Improvements and testing needed.

Changed

  • Moved group_columns method, series_type, and type definitions from the capdata module to the columngroups module and changed group_columns from a CapData method to a function that returns a ColumnGroups instance.

  • Completely refactored the algorithim to determine the reporting irradiance when irr_bal is set to True. Now an instance of the new ReportingIrradiance class is created and used to determine the reporting irradiance.

  • Updated the examples to reflect the changes to the API for data loading, column grouping, selection (loc and floc vs view and rview), and balanced reporting irradiance.

  • Changed the scatter_hv method to be more flexible and not require temperature or wind regression columns.

Removed

  • Removed the value checking functionality of the group_columns function. Data quality is outside the scope of the pvcaptest project.

  • Removed the load_data, load_das, and load_pvsyst methods from the CapData class. Replaced with io module.

0.10.0 - 2021-07-25

Added

  • Added the filter_missing CapData method to remove missing data from specified columns. By default removes only intervals that contain missing data in the regression variable columns.

  • Added option to filter_irr method to specify using the reporting irradiance in the CapData object as the reference irradiance.

  • Added option to the filter_time method to drop the specified time period instead of dropping all other times.

  • Added option to filter_clearsky method to keep time periods with unstable irradiance.

  • Added new attributes to CapData: removed, kept, filter_counts. The update_summary decorator now stores the index of points removed, the index of points remaining after each filter, and the number of times any filter has been run for each filter applied.

  • Adds new plotting method, scatter_filters, which shows which filtering step removed which time intervals of data in a plot of irradiance vs. power.

  • New plotting method, timeseries_filters, which shows which fitlering step removed which time intervals of data in a plot of power vs. time.

  • New plotting function, overlay_scatters, that overlays irradiance vs. power scatter plots of the data remaining after the last filtering step of the two CapData objects passed to the function.

  • New get_filtering_table method that returns a DataFrame documenting the which time intervals are removed by which filter and which time intervals remain after all filtering.

  • Adds the run_test function, which applies the passed list of CapData filtering methods to the CapData object passed.

  • Adds the points_summary method, which prints the number of points remaining after all filtering, the length of the test period, the average points remaining after filtering per day, if enough points have been collected, if more points are needed how many, and how many days left if the rate of points holds.

Changed

  • Updated filter_pvsyst method to handle inverter output variables that have underscores or spaces like ‘IL Pmin’ and ‘IL_Pmin’.

  • load_das method no longer drops columns and rows that contain no data

  • Format of hover tooltip in plots produced by plot method now includes comma separator for thousands.

  • Changes captest to pvcaptest in documentation.

  • get_reg_cols method default changed to get and rename the columns defined in the regression_cols attribute rather than expecting regression variables/columns to be identified by the keys ‘power’, ‘poa’, ‘t_amb’, and ‘w_vel’ in the regression_cols attribute.

0.9.0 - 2020-08-16

Changed

  • Updated clear sky functions which rely on pandas index.tz_localize to use nonexistent argument rather than errors argument, which was deprecated in pandas v1.0.

  • Made Pandas v1.0 or greater a requirement for pvcaptest.

  • Change to test against python v3.7* and v3.8*

  • Updated installation instructions.

0.8.0 - 2020-04-13

Added

  • Added a filter_power method to the CapData class.

  • Added a filter_days method to the CapData class.

Changed

  • Allow get_reg_cols to accept a single regression variable as a string. Previously required passing list with at least two entries.

  • Fixed bug in filter_clearksy that applied filter to data rather than data_filtered attribute.

  • Added option to plot method to use column names for hover labels instead of abbreviated column names.

  • Improved formatting of the filtering summary output. See issue #12 for details.

  • Cleaned up source code by correcting linter errors.

0.7.0 - 2020-03-08

Added

  • New filter_shade method separate from the filter_pvsyst method.

  • captest_results method warns when it automatically attempts to correct for W vs kW.

Changed

  • Filter_pvsyst method filters on IL Pmin, IL Pmax, IL Vmin, and IL Vmax and warns if any of the four are missing. Previously failed if any of the four were missing.

  • cp_results returns a warning if the regression formulas of the passed CapData objects do not match instead of warning and continuing.

  • Updates to make captest compatible with pvlib 0.7.0

  • Editing of the complete capacity test example to use new names and improve explanations of features.

Names were changed to remove ambiguous abbreviations:

  • flt - filter; API changes in many places

  • cntg_eoy - wrap_year_end; API change

  • cp_results - captest_results; API change

  • res_summary - captest_results_check_pvalues; API change

  • reg_fml - regression_formula; API change

  • irrRC_balanced - irr_rc_balanced; API change

  • df_beg - df_start

  • ix_ser - ix_series

  • mnth - month

  • months_boy - months_year_start

  • months_eoy - months_year_end

  • loop_cnt - loop_count

  • cprat - cap_ratio

  • cprat_cpval - cap_ratio_check_pvalues

  • trans - column_groups; API change

  • set_translation - group_columns

  • trans_report - column_type_report

  • set_trans argument of load_data - group_columns

  • review_trans - review_column_groups

  • set_reg_trans - set_regression_cols

  • reg_trans - regression_cols

  • update_reg_trans argument of agg_sensors - update_regression_cols

  • reg_cpt - fit_regression

  • ols_model - regression_results

Removed

  • Removed the inv_trans_dict function. This was intended for use within the module and was unused.

0.6.0 - 2019-09-15

Added

  • Setup Travis CI to test pull requests and test and deploy to pypi for tags on master.

  • Setup project documentation hosted by Read the Docs using sphinx, nbshpinx, napolean, recommonmark, AutoStructify

Changed

  • Versioning changed from manual update in __version.py file to using versioneer to update version number from git tag.

  • Updated this file to follow the Keep a Changelog formatting conventions.

  • Moved repository to an organization github account from my personal github account.

  • Examples moved from root/examples directory to docs/examples. Executed versions of the examples display on read the docs.

  • All examples can be launched through binder in live notebooks.

  • The environment file has been updated to work for binder and Read the Docs.

0.5.3 - 2019-05-12

Changed

  • Update name and location of conda environment yml file, so there is a single file and it works with binder.

  • Removed binder directory.

  • Update readme to reflect changes to conda environment.yml

  • Minor updates to example.

  • Minor documentation string updates.

0.5.1 - 2019-05-01

Added

  • Addition of clear sky modeling using pvlib library. See new example notebook ‘Clear Sky Examples’.

  • Added a new method, predict_capacities for calculating reporting conditions and predicted outputs by month.

  • New example notebook demonstrating use of rep_cond and predict_capacities.

  • Add warning when filter removes all data.

Changed

  • Changed Holoviews dependency to >= v1.11. DatLink added in v1.11 is required for scatter_hv method.

  • Expanded docstring for the load_data method to more clearly explain how the method joins multiple files (by row).

  • Update installation directions in README.

  • Updated conda environment file (conda_env.yml) to match updated dependencies.

  • Moved all filtering and regression functionality from CapTest class into the CapData class and replace CapTest class with functions for results comparing CapData objects.

  • Significant refactor of the rep_cond function. Removed any time filtering and prediction from rep_cond. Rep_cond acts on filtered data in the df_flt attribute.

  • agg_sensors method updated to be more explicit and flexible.

  • Changed filter_sensors to filter based on percent difference between all combinations of pairs of sensors measuring the same environmental factor. Corrected bug where standard deviation filter could not detect outliers with more than two, but still a small number of sensors.

  • Adjusted bounds check of columns of data when importing so that translation dictionary names would not have ‘valuesError’ added to them.

  • Made printout of bounds check results optional when loading data.

  • Adjusted the type_defs and sub_type_defs, so that translation dictionary keys are more accurate for PVsyst data.