test: run the balance-slack tests on both backends - #8
Open
PabloBotin wants to merge 11 commits into
Open
Conversation
Pin the current fuel/demand data contract ahead of the PowerAnalytics metrics-API migration: category naming (In/Out split, Curtailment, slack display names), charging sign conventions, palette-first column ordering, demand column naming, time-window and filter_func kwargs, and per-backend series counts.
PowerAnalytics imports get_system from PowerSimulations, so the unexported PA.PSI alias is unnecessary. Also extend the missing-system error to mention loading results with populate_system = true.
The IS.Results path now computes Metrics.calc_load_forecast over the all_loads selector (grouped into a single column renamed to "Load" so palette and label behavior are unchanged); a user filter_func folds into the selector. Time windows (initial_time/horizon, also spelled start_time/len) are applied by local row slicing because compute rejects unknown kwargs and mishandles len on simulation results in PA 1.4. The PSY.System path stays on the old get_load_data API, which has no new-API equivalent. The dead isnothing guard on the aggregated demand frame is replaced by an isempty check that can actually fire.
Assemble the fuel stack from PowerAnalytics Metric/ComponentSelector primitives instead of get_generation_data/make_fuel_dictionary/ categorize_data/combine_categories, preserving the exact column set, order, names, and signs. Components are assigned to a single category by replaying the old first-match-wins priority over the per-rule subselectors (the independent new selectors would otherwise double-count, e.g. NG-CC vs NG-Steam); generators fall back variable -> forecast parameter -> PowerOutput aux; storage/sources split into '<category> In'/'<category> Out' with charging flipped negative; slacks keep their BALANCE_SLACKVARS display names; unmatched components go to 'Other' with an error log. Also fix the net-load overlay to actually include storage charging by passing the charging total as extra_load, update the test mapping yaml for the new parser's strict fuel enums, and pin both behaviors with new tests.
…ta(::PowerData) plot_results now owns its dict-of-DataFrames path (DateTime stripped per entry, time axis from the first entry) instead of constructing PowerAnalytics.PowerData. The plot_powerdata methods move to src/deprecated.jl as forwarding shims that warn about removal in a future breaking release. combine_categories = false no longer crashes: it plots one trace per stored column, and the docstrings now state the actual default (true).
The Weave report template's tables now use the PowerAnalytics metrics API (calc_active_power per fuel category, calc_system_load_forecast) instead of the deprecated get_generation_data/get_load_data accessors; the Services table is dropped since get_service_data has no metrics-API equivalent. Docstrings drop the never-functional plot_fuel 'variables' kwarg, document the storage/sources kwargs, and reference plot_results instead of the deprecated plot_powerdata. The public API reference gains a hand-written Deprecated section.
- plot_demand no longer crashes when a load type has no results; missing
results skip to the "No load data found" path (now an ArgumentError)
- warn on the unsupported `variables` kwarg of plot_fuel instead of
silently ignoring it
- warn when a custom generator mapping yaml contains ext_category keys,
which the PowerAnalytics 1.0 selector parser cannot honor
- _combine_result_categories: unknown `names` entries raise an actionable
ArgumentError; Vector{Symbol} accepted for the deprecated powerdata path
- docstrings: aggregate scope (System path only), time-window kwargs and
aliases, aggregate-function return-shape contract
- _FuelRule stores type_name::Symbol to avoid per-supertype allocations
- test: old-vs-new numeric equivalence of fuel category traces
…names
Every plot function now takes `backend::PlottingBackend`, defaulting to
`CairoMakieBackend()`:
plot_fuel(res) # CairoMakie
plot_fuel(res; backend = PlotlyLightBackend()) # PlotlyLight
The backend was already modeled as a value in src/backends.jl, so encoding it in
the function name doubled the public API without buying any dispatch. The ten
`_plotly`-suffixed functions keep working but warn and forward. Passing both a
`_plotly` name and a `backend` key word raises an ArgumentError rather than
letting one silently win, since the two would disagree about the renderer.
Eight per-plot behaviors that were resolved twice, once in each recipe, are now
resolved once in call_plots.jl and handed to the recipes through _PlotOptions:
fill default, line width, line style, draw order, title sentinel, empty input,
save path, and palette selection.
User-visible changes:
- Both backends select from the whole palette returned by `load_palette`, so
more series get a distinct color before the cycle repeats. PlotlyLight
previously drew from a narrower set, so its default colors change.
- `_default_save_format` dispatches on the backend, making the PlotlyLight
default `html`. A shared hardcoded "png" tripped the extension-rewrite warning
on every default-path PlotlyLight save. An explicit `format` still wins.
- CairoMakie non-stacked draw order now matches PlotlyLight.
- WeaveExt = ["PlotlyLight", "Weave"] -> "Weave". Neither the extension nor
generic_report_template.jmd touches PlotlyLight, so the old trigger withheld
`report` from a CairoMakie-only user.
The backend stubs dispatch per concrete backend. With `backend` defaulting to
CairoMakie, a PlotlyLight-only user reached a stub telling them to run `using
PlotlyLight` when they already had; each stub now names its own package, and the
CairoMakie one names the key word that selects the other backend.
Two save-path defects go with it. `_resolve_save_file` is now the single place a
save path is decided, and it replaces spaces in the title with underscores as
every entry point on main already did; centralizing the path had dropped that
for `plot_dataframe` alone. `_plot_demand!` read `:save` without removing it
from the key words it forwarded, so one call saved the figure twice under two
different names; it now strips `:save`, `:title`, and `:set_display` like the
other wrappers.
Tests: plot_introspection.jl reads rendered marks back out of both libraries so
value assertions run against either backend; test_backend_parity.jl enforces the
parity contract; test_demand_semantics.jl and test_fuel_categories.jl pin the
demand sign and the fuel-rule specificity. Suite is at 368 pass / 0 fail.
Docs gain explanation/backend_parity.md and a Change Backends how-to; the
orphaned explanation/stub.md is removed. Personal notes are ignored through the
user-level git ignore rather than this repository's shared .gitignore.
Route the report template's Load table through the new public `get_demand_data` rather than `calc_system_load_forecast`, which reported the requested instead of the served demand and disagreed with `plot_demand` under controllable load formulations. Delegate `_combine_result_categories` to `PowerAnalytics.combine_categories` instead of reimplementing it, keeping only the actionable error on an unknown `names` entry. Move `seriescolor`, `column_labels`, `interval`, the scaled data matrix and the net-sign classification into `_PlotOptions`, so neither recipe derives them independently and the third spelling of the sign test disappears. Lowercase the extension in the PlotlyLight writer so `.HTML` is recognized rather than silently rewritten to a different path, and pin it with a test. Hoist the shared "Accepted Key Words" documentation into `_COMMON_PLOT_KWARGS` and interpolate it, replacing eight verbatim copies. Keep `_report_plot_fuel` as a forwarding shim: report templates copied from an earlier release call it positionally, so removing it would throw `UndefVarError` on their next `report`.
Slack categories were read through `PA.Metrics.calc_system_slack_up` and its down counterpart, which resolve to `PowerAnalytics.read_system_result`. That function hardcodes `make_key(entry, PSY.System)` and asserts the result has a single data column. PowerSimulations attaches balance slacks to four different component types depending on the network formulation: `System` for copper-plate and PTDF, `Area` for the area-balance models, and `ACBus` for the PowerModels formulations (with `"P"`/`"Q"` meta in the AC case). Only the first was ever found, so node- and area-level slacks silently never appeared — and would have tripped the single-column assertion had they been read at all. Discover the slack keys present in the results by entry type instead, regardless of which component owns them, and sum across the per-bus or per-area columns into one series per direction. The `System` case has a single column, so its values are unchanged. Only the active-power `"P"` slack is included for AC formulations; the fuel stack plots active power, so reactive slack does not belong on it. This works around the limitation rather than fixing it: `read_system_result` is still `System`-only for every other PowerAnalytics consumer.
The Sienna-Platform#94 tests read traces off `plot_fuel_plotly`'s PlotlyLight plot object directly, which pinned them to one backend — the same asymmetry that let the bar-stacking fix in Sienna-Platform#140 land in one recipe and not the other. They now loop over `FUEL_BACKENDS` and read through the `series_values`/`series_labels` helpers, so both backends are checked. The oracles are computed once outside the backend loop and the shared `fuel_results_ed`/`fuel_results_uc` fixtures replace a second solve, so the added coverage costs no extra optimization run.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Runs the balance-slack tests from Sienna-Platform#94 on both plotting backends instead of PlotlyLight only.
The tests read traces off
plot_fuel_plotly's plot object directly, which pinned them to onebackend. That is the same asymmetry that let the bar-stacking fix in
#140 land in one recipe and
silently not the other. They now loop over
FUEL_BACKENDSand read through theseries_values/series_labelshelpers fromplot_introspection.jl, so both backends areasserted against the same oracle.
The oracles are computed once outside the backend loop, and the shared
fuel_results_ed/fuel_results_ucfixtures replace a second solve — so the added coverage costs no extraoptimization run.
julia --project=test test/runtests.jl test_fuel_stack_behavior→ 130 pass / 0 fail.Note on the diff size
This branch is stacked on
feature/migrate-to-poweranalytics-new-api(the PowerAnalytics 1.0migration, PR Sienna-Platform#145 upstream) and on
issue-94-bus-slacks(fork PR #7). Because it targetsmain, the diff shown here includes all of those commits as well.Only the last commit,
08a4387, is new here — a single file,test/test_fuel_stack_behavior.jl.Everything before it is under review elsewhere.