Changelog

Development

Placeholder

4.0.7

Handle ambiguous and nonexistent local times when creating daily dataclass

4.0.6

Update docs.
Update typehints on core daily and utility functions.
Minor change to loading test data to ensure the reporting period is a year ahead of the baseline period.

4.0.5

Flip slope when deserializing legacy hdd_only models

4.0.4

Add support for deserializing legacy hourly models
Fix legacy daily model deserialization

4.0.3

Move masking behavior for rows with missing temperature from reporting dataclass to prediction output
Add disqualification check to billing model predict()

4.0.2

Force index to use nanosecond precision
Compute coverage using same offset as initial reads to fix issues when downsampling hourly data
Update test data location
Fix bug in daily plotting to remove NaN values if input
Refactor sufficiency criteria to be more explicit and easier to manage

4.0.1

Correct dataframe input behavior and final row temperature aggregation
Remove unnecessary datetime normalization in order to respect hour of day
Convert timestamps in certain warnings to strings to allow serialization
Allow configuration of segment_type in HourlyModel wrapper

4.0.0

Update daily model methods, API, and serialization
Provide new API for hourly model to match daily syntax and prepare for future additions
Add baseline and reporting dataclasses to support compliant initialization of meter and temperature data

3.2.0

Addition of modules and amendments in support of international facility for EEMeter, including principally:
Addition of quickstart.py; updating setup.py and init/py accordingly.
Inclusion of temperature conversion amendments to design_matrices; features; and derivatives.
Addition of new tests and samples.
Amendments to tutorial.ipynb.
Addition of eemeter international.ipynb.
Change .iteritems() to .items() in accordance with pandas>=2.0.0
.get_loc(x, method=...) to .get_indexer([x],method=...)[0] in accordance with pandas>=2.0.0
Updated mean() to mean(numeric_only=True) in accordance to pandas>=2.0.0
Updated tests to work with pandas>=2.0.0
Update python version in Dockerfile.
Update other dependencies (including adding rust) in Dockerfile.
Remove pinned dependencies in Pipfile.
Relock Pipfile (and do so inside of the docker image).
Update pytests to account for changes in newer pandas where categorical variables are no longer included in df.sum().sum().
Clarify the functioning of start, end and max_days parameters to get_reporting_data() and get_baseline_data().

3.1.1

Update observed_mean calculation to account for solar (negative usage) to provide sensible cvrmse calculations.

3.1.0

Remove missing hour_of_week categories in the CalTrack hourly methods so they predict null for those hours.

3.0.0

Remove python27 support.
Update Pipfile lock.
Update fit_temperature_bins to potentially take an occupancy_lookup in order to fit different temperature bins for occupied/unoccupied modes. This changes the args passed to eemeter.create_caltrack_hourly_segmented_design_matrices, where it now requires a set of bins for occupied and unoccupied temperatures separately.
Update CalTRACK hourly model formula to use different bins for occupied and unoccupied mode.

2.10.11

Fix tests and make changes to ensure tests pass on pandas version 1.2.1.
Fix bug in segmentation.py causing a section of tutorial to fail.

2.10.0

Add additional terms into ModelMetrics() class which can be used in fractional savings uncertainy computations.

2.9.2

Remove fixing of versions of libraries in setup.py to avoid unforeseen issues with library updates.

2.9.1

Fix versions of libraries in setup.py to avoid unforeseen issues with library updates.

2.9.0

Clarify blackout period.

2.8.6

Fix issue with get_reporting_data and get_baseline_data when passing data with non-UTC timezones.

2.8.5

Add functions to clean billing/daily data according to caltrack rules.

2.8.4

Further limit segments used in hourly totals_metrics to only calculate when weight=1.

2.8.3

Update hourly totals_metrics calculation to properly use only the segment of the model.

2.8.2

Add totals_metrics to hourly models.

2.8.1

Fix bug with get_baseline_data in regards to recent addition of n_days_billing_period_overshoot kwarg.

2.8.0

Update get_baseline_data to allow for limit to billing overshoot using n_days_billing_period_overshoot kwarg.

2.7.7

Add function to clean billing data to fit caltrack specifications (clean_caltrack_billing_data).

2.7.6

Update io functions to support latest pandas (>=0.24.x).
Update documentation for CalTRACK Hourly methods.
Add tutorial.

2.7.5

Fix completeness check for get_terms for last term.

2.7.4

Make more usable outputs for the get_terms function (list of eemeter.Term objects).

2.7.3

Update as_freq so it has an optional include_coverage parameter where it returns a dataframe with one column including the percent coverage of data used to create each sample.

2.7.2

Fixes the columns that are given in an empty prediction result called with the with_design_matrix=True flag set for caltrack usage per day methods.
Update bug report github issue template.
Add test for as_freq.

2.7.1

Change as_freq to handle all Null series.

2.7.0

Add get_terms method to allow splitting reporting data into any number of terms specified by day length.

2.6.0

Change fit_caltrack_hourly_model so it returns a CalTRACKHourlyModelResults object rather than a CalTRACKHourlyModel, in order to bring it in line with the caltrack_usage_per_day model outputs.

2.5.4-post1

Update MANIFEST.in to fix release and update ./bump_version.sh script to remove build directories.

2.5.4

Add data fields to the DataSufficiency even if there are no warnings when calculating sufficiency.

2.5.3-post2

Attempt 2 to fix release .whl file by removing local build and dist directories before running python setup.py upload.

2.5.3-post1

Fix release .whl file which had some extra directories.
Add draft MAINTAINERS.md.

2.5.3

Fix metered_savings behavior so that it does not fail to compute error bands when there is 0 variance in the baseline.

2.5.2

Fix as_freq behavior to preserve sum and add a null last index at the target frequency if necessary.

2.5.1

Capture an additional exception type (KeyError) in recently adjusted get_baseline_data and get_reporting_data methods.

2.5.0

Add parameters to get_baseline_data and get_reporting_data to help make these methods a bit more correct for billing data.
Preserve nulls properly in as_freq.
Update jupyter version to be compatible with latest tornado version.

2.4.0

Fix for bug that occasionally leads to LinAlgError: SVD did not converge error when fitting caltrack hourly models by addressing multi-collinearity when only a single occupancy mode is detected

2.3.1

Hot fix for bug that occasionally leads to LinAlgError: SVD did not converge error when fitting caltrack hourly models by converting the weights from np.float64 ton np.float32.

2.3.0

Fix bug where the model prediction includes features in the last row that should be null.
Fix in transform.get_baseline_data and transform.get_reporting_data to enable pulling a full year of data even with irregular billing periods

2.2.10

Added option in transform.as_freq to handle instantaneous data such as temperature and other weather variables.

2.2.9

Predict with empty formula now returns NaNs.

2.2.8

Update compute_occupancy_feature so it can handle instances where there are less than 168 values in the data.

2.2.7

SegmentModel becomes CalTRACKSegmentModel, which includes a hard-coded check that the same hours of week are in the model fit parameters and the prediction design matrix.

2.2.6

Reverts small data bug fix.

2.2.5

Fix bug with small data (1<week) for hourly occupancy feature calculation.
Bump dev eeweather version.
Add bump_version script.
Filter two specific warnings when running tests: statsmodels pandas .ix warning, and eemeter model fitting warning.

2.2.4

Add json() serialization for SegmentModel and SegmentedModel.

2.2.3

Change max_value to float so that it can be json serialized even if the input is int64s.

2.2.2

Add warning to caltrack_sufficiency_criteria regarding extreme values.

2.2.1

Fix bug in fractional savings uncertainty calculations using billing data.

2.2.0

Add fractional savings uncertainty to modeled savings derivatives.

2.1.8

Update so that models built with empty temperature data won't result in error.

2.1.7

Update so that models built from a single record won't result in error.

2.1.6

Update multiple places where df.empty is used and replaced with df.dropna().empty.
Update documentation for running CalTRACK hourly methods.

2.1.5

Fix zero division error in metrics calculation for several metrics that would otherwise cause division by zero errors in fsu_error_band calculation.

2.1.4

Fix zero division error in metrics calculation for series of length 1.

2.1.3

Fix bug related to caltrack billing design matrix creation during empty temperature traces.

2.1.2

Add automatic t-stat computation for metered savings error bands, the implementation of which requires expicitly adding scipy to setup.py requirements.
Don't compute error bands if reporting period data is empty for metered savings.

2.1.1

Fix degree day ranges (30-90) for prefab caltrack design matrix creation methods.
Fix the warning for total degree days to use total degree days instead of average degree days.

2.1.0

Update the use_billing_presets option in fit_caltrack_usage_per_day_model to use a minimum data sufficiency requirement for qualifying CandidateModels (similar to daily methods).
Add an error when attempting to use billing presets without passing a weights column to facilitate weighted least squares.

2.0.5

Give better error for duplicated meter index in compute temperature features.

2.0.4

Change metrics input length error to warning.

2.0.3

Apply black code style for easy opinionated PEP 008 formatting
Apply JSON-safe float conversion to all metrics.

2.0.2

Cont. fixing JSON representation of NaN values

2.0.1

Fixed JSON representation of model classes

2.0.0

Initial release of 2.x.x series