Switch 2.0.6 #115
Switch 2.0.6 #115
Conversation
|
@rodrigomha and @josiahjohnston I added a commit with simplified code for the rhosetter in the PySP example, with no dependency on Since this is your code, would you be able to see what you can do to get it to work? (I'm also thinking we should have a bigger test set that we run before each release to make sure all our examples (and some bigger ones) run and produce the same results as the previous release.) Here's the command I used that gave an error: Here were the pdb commands I used to diagnose the error (a little): |
DependenciesRetiring the inefficient & roundabout sympy parsing in favor of direct Pyomo calls seems great. Dropping the direct Consolidating the rest of the optional dependencies into a single section called I'm not a fan of making users do ad-hoc installations, and I don't understand benefits of doing that vs registering them explicitly as optional bits in My preference is to keep testfixtures in the core requirements since the testing framework is an integral part of software development, and at this point Switch still seems best-suited for people can do some coding. Hopefully Switch will grow into a broader audience, but in the meanwhile, it seems important to make it easy for people to start developing Switch to expand the base of advanced users & contributors. Moving testfixtures into Maybe you can elaborate on the motivation for reducing dependencies? Design esthetics or some practical problems? InstallationsThe instructions for casual users seems fine to me. For advanced users who will be working with code, I strongly recommend using some kind of virtual environment. Conda's environments seem fine as far as I know, as is virtualenv. I started mucking with pipenv (which seems to be taking over from virtualenv), but I haven't gotten far enough with pipenv to recommend it or write instructions. I always recommend using I've played with the idea of setting up a docker image for Switch, but didn't know of a concrete use case. If that ever seems of interest to anyone, I could do that pretty easily. Source code accessI like your idea of I'm not sure how or if the Adding something like rhosetter & Progressive HedgingThe problem was that the input .dat files needed to be regenerated since |
|
I reviewed the PySP example and fixed several bugs, as well as updated the output files to match the format of PySP's current version. I too am getting objective function values of about 135 MMUSD. I checked the output files and manually estimated the value of the OF and got closer to this value than to the previous ~110. I suspect that we hadn't re-generated the input files and re-run the example since I first uploaded it, so any changes to Switch's OF hadn't been reflected in the output files of this example. I vote to keep these updated output files and consider them to be correct. |
|
Thanks Benjamin! |
|
Just pushed some more updates that pass all tests in both python 2 & 3:
|
… extra timepoints warnings to avoid spamming user with messages at every run, but giving them access to as detailed information as they request.
…n, and easier to read on a laptop screen (adding `dest` and reformatting text wrap).
… were built before the start of the study's time horizon.
…elp messages. Before this fix, the bug appeared if you executed `switch solve --help` in the examples/production_cost_models/spinning_reserves directory which loads the spinning_reserves module.
…hon library `ggplot` has been abandoned for a few years, and `plotnine` is actively maintained and a drop-in replace for our simple use case.
…le. This allows tab completion & color highlighting, which enables much faster exploration & debugging.
…s warnings about matplotlib API deprecations & save paths for plotnine figures.
…()` (autoselect vs auto_select)
…ants used in high-flow conditions where river flow exceeds the capacity of the generator & reservoir. Also minor improvements to documentation and code formatting for improved readability & accuracy.
* New file with annual summary per-generation project that includes energy, capacity, costs & capacity factor * New file with tech/annual summary for easily skimming results. * Faster summation of tech/zone annual summary & tech/annual summary. Also include capacity, costs, and cap factor in those summary files. Also, minor updates for python style.
…t some parameters are optional in the model and don't have to be specified in files. This simplifies the process of defining a bare-bones planning reserve margin for each load zone. I verified that all of the tests still pass.
…tion fuel costs associated with dispatching generators. Also, reverting a typo bug in dispatch export.
…d to dispatch with the fuel markets module.
…d in input files, so the implementation matches documentation.
…nts that are retired early and plants whose construction is not complete in the first period. Also minor documentation update.
…thias made in mid-June 2019.
…ng.__init__ This ensures that users can prevent all reporting by removing the reporting module from the module list and avoids a crash if they tried to do that previously (due to the gen_cap code trying to access model.options.sort_output, which is defined in reporting.__init__).
Pyomo 5.7 made several changes that broke compatibility with earlier models:
- stopped assuming dimen values for sets when they were not specified, which
crashed some of our input and output code.
- began preserving order in all Sets and emitting warnings if unordered
containers such as standard Python sets are used to initialize them (this was
done to improve determinism in the model setup).
- +inf, -inf, +infinity, -infinity and nan in the input files are now read in as
strings instead of floating point values.
- a deprecation warning is issued saying that the default domain for Params will
be Reals instead of Any, starting in version 6.0.
- "Simple" has been changed to "Scalar" in the names of SimpleSet,
OrderedSimpleSet, AbstractOrderedSimpleSet and SimpleParam.
Pyomo 6.0 also began producing deprecation warnings when itervalues or iteritems
methods are used on Pyomo components.
To make Switch compatible with Pyomo 5.7 - 6.4, this commit makes the following
changes:
- assign dimen values to all sets (preferably as the first keyword argument)
- use lists instead of sets to initialize Set components
(switch_model.utilities.unique_list(iter) is helpful for this.)
- replace "inf" in some fuel_supply_curves.csv example files with ".", which
will cause them to use the default value, which is float("inf").
- don't specify rfm_supply_tier_limit when the fuel_costs.markets module reads
additional fuels from fuel_costs.csv. They get a default value of float("inf")
anyway.
- add a note to CHANGELOG.md that using "inf" and "nan" values in the input
files will generate errors with Pyomo 5.7+.
- set all Params to have an explicit domain
- also check for ScalarSet, OrderedScalarSet, AbstractOrderedScalarSet and
ScalarParam in the locations that check for SimpleSet, OrderedSimpleSet,
AbstractOrderedSimpleSet and SimpleParam.
…enerators.post_solve
This commit applies --sorted-output automatically in reporting.write_tables and explicitly in several locations where pandas dataframes are written to .csv files. This ensures most outputs will be sorted when requested, including dispatch.csv, dispatch-wide.csv (columns), gen_cap.csv, gen_project_annual_summary.csv, load_balance.csv, local_td_energy_balance.csv, local_td_energy_balance_wide.csv and transmission.csv.
…predetermined.csv This also renames gen_predetermined_storage_energy_mwh (new in this release, 2.0.7) to build_gen_energy_predetermined. The new names should be more consistent with each other and also be a little clearer that this is the amount of new capacity built in each year, not the total amount of capacity in place that year.
The new name is more consistent with other file names (which generally start with "gen_"), and should be a little easier to remember and view in editor tabs.
…td_loss_rate This change encapsulates the local_td data more completely within the local_td module and also gives the parameter a more accurate name and allows it to be varied between load_zones.
…e components Both of these changes are needed for Pyomo 6+ compatibility. Pyomo issues a deprecation warning for Params with default "Any" domain, due to future plans to change this to "Reals". It also fails to clone models that use obj.keys() to initialize components (probably not able to clone/pickle the iterator).
… members `build` is a Python variable created as the sum of several Pyomo components. We previously checked whether it was empty via `build is 0`, but Python now gives a warning about `is` with a literal, so we changed it to `build == 0`, but that forces evaluation of an unconstructed component, which crashes Pyomo. So instead we check `isinstance(build, int) and build == 0`. The second part is not actually needed, but clarifies what is happening.
To do this, I had to move the banner into SwitchAbstractModel.__init__ between the parameter parsing code and the define_components code (or else split __init__ into multiple methods that have to be called in the right order from switch_model.solve). To do that, I had to make iteration_modules a part of the model instead of a temporary variable in switch_model.solve, which seems reasonable anyway. This commit also converts a number of print() calls into logger.info() calls.
…der solve-scenarios
fix typo in planning_reserves and split storage_builds.csv into storage_builds.csv and storage_capacity.csv
…age values This commit changes gen_cap.csv to include both power and energy (storage) capacity in place in each period. gen_cap.csv also shows the annual capital recovery and fixed O&M required for each project. This commit also adds a gen_build.csv file that shows the amount of power and energy capacity _added_ in each period and the total capital outlay needed when those are built. (Switch uses capital outlay to calculate capital recovery requirements, but doesn't outlay the capital outlay directly. However, capital outlay is of interest to some users.)
|
Closing this because it was actually the 2.0.6 branch, not the next_release branch. |
(Note: most of this comment and Josiah's response have been moved to issue #129 for future implementation.)
This is a place to discuss the next release of Switch. These are my immediate goals:
conda install switch_model_extras/pip install switch_model[extras])conda install -c defaults -c conda-forge switch_modelorpip install switch_model(+find glpk somewhere) for most users?git clone https://github.com/switch-model/switch.git && cd switchconda install --only-deps switch_modelpip install --editable .orpython setup.py develop(note:conda developdoesn't install command-line scripts or dependencies; see how to use conda develop? conda/conda-build#1992 (comment))switch find <module>: report file path to specified module (possibly just a submodule within switch_model)atom `switch find switch_model`,mate `switch find pyomo`or maybeatom `switch find discrete_commit`.switch install examples: copy examples directory to local directoryswitch solve --trace [[<module>|<file>[:<function>|<line>]], [<module>[:<function>]], ...]: invoke the debugger (a) when particular callbacks are called, (b) when any callback in the specified module is called (if no function specified), or (c) whenever any callback is called (if no modules specified).