Source code for message_ix_models.model.transport.report

"""Reporting/postprocessing for MESSAGEix-Transport."""

import logging
from copy import deepcopy
from pathlib import Path
from typing import TYPE_CHECKING, Any, Tuple

import genno
import pandas as pd
from genno import Computer, Key, KeySeq, MissingKeyError
from genno.core.key import single_key
from message_ix import Reporter

from message_ix_models import Context, ScenarioInfo
from message_ix_models.report.util import add_replacements

from . import Config

if TYPE_CHECKING:
    import ixmp
    from genno import Computer
    from genno.types import AnyQuantity

    from message_ix_models import Spec

log = logging.getLogger(__name__)


[docs]def check(scenario): """Check that the transport model solution is complete. Parameters ---------- scenario : message_ix.Scenario Scenario with solution. Returns ------- pd.Series Index entries are str descriptions of checks. Values are :obj:`True` if the respective check passes. """ # NB this is here to avoid circular imports from message_ix_models.report import prepare_reporter, register register(callback) rep, key = prepare_reporter(scenario, "global.yaml", "transport check") return rep.get(key)
[docs]def aggregate(c: "Computer") -> None: """Aggregate individual transport technologies to modes.""" from genno.operator import aggregate as func config: Config = c.graph["config"]["transport"] for key in map(lambda s: KeySeq(c.infer_keys(s)), "emi in out".split()): try: # Reference the function to avoid the genno magic which would treat as sum() # NB aggregation on the nl dimension *could* come first, but this can use a # lot of memory when applied to e.g. out:*: for a full global model. c.add(key[0], func, key.base, "t::transport agg", keep=False) c.add(key[1], func, key[0], "nl::world agg", keep=False) c.add(key["transport"], "select", key[1], "t::transport modes 1", sums=True) except MissingKeyError: if config.with_solution: raise
[docs]def reapply_units(c: "Computer") -> None: """Apply units to transport quantities. :func:`.ixmp.report.operator.data_for_quantity` drops units for most data extracted from a MESSAGEix-GLOBIOM :class:`.Scenario`, because the data contain a mix of inconsistent units. Here, add tasks to reapply units to selected subsets of data that are guaranteed to have certain units. """ # TODO Infer these values from technology.yaml etc. for base, (op, units) in { # Vehicle stocks # FIXME should not need the extra [vehicle] in the numerator "CAP:nl-t-ya:non-ldv": ("apply", "v**2 Tm / a"), "CAP:*:ldv": ("apply", "Mv"), "CAP_NEW:*:ldv": ("apply", "Mv"), # NB these units are correct for final energy only "in:*:transport": ("apply", "GWa / a"), "in:*:ldv": ("apply", "GWa / a"), "out:*:transport": ("apply", "Tm / a"), "out:*:ldv": ("apply", "Tm / a"), # Units of ACT are not carried, so must correct here: # - Add [time]: -1 # - Remove [vehicle]: -1, [distance]: -1 # # When run together with global.yaml reporting, emi:* is assigned units of # "Mt / year". Using apply_units() causes these to be *converted* to kt/a, i.e. # increasing the magnitude; so use assign_units() instead. "emi:*:transport": ("assign", "kt / a"), }.items(): key = c.infer_keys(base) c.add(key + "units", f"{op}_units", key, units=units, sums=True)
SELECT = [ "CAP_NEW", "CAP", "fix_cost", "historical_new_capacity", "in", "input", "inv_cost", "out", "var_cost", ]
[docs]def select_transport_techs(c: "Computer") -> None: """Select subsets of transport technologies.""" # Infer the full dimensionality of each key to be selected for key in map(lambda name: c.infer_keys(f"{name}:*"), SELECT): c.add(key + "transport all", "select", key, "t::transport all", sums=True) c.add(key + "ldv", "select", key, "t::transport LDV", sums=True) c.add(key + "non-ldv", "select", key, "t::transport non-ldv", sums=True)
# TODO Type c as (string) "Computer" once genno supports this
[docs]def add_iamc_store_write(c: Computer, base_key) -> "Key": """Write `base_key` to CSV, XLSX, and/or both; and/or store on "scenario". If `base_key` is, for instance, "foo::iamc", this function adds the following keys: - "foo::iamc+all": both of: - "foo::iamc+file": both of: - "foo::iamc+csv": write the data in `base_key` to a file named :file:`foo.csv`. - "foo::iamc+xlsx": write the data in `base_key` to a file named :file:`foo.xlsx`. The files are created in a subdirectory using :func:`make_output_path`—that is, including a path component given by the scenario URL. - "foo::iamc+store" store the data in `base_key` as time series data on the scenario identified by the key "scenario". .. todo:: Move upstream, to :mod:`message_ix_models`. """ k = KeySeq(base_key) file_keys = [] for suffix in ("csv", "xlsx"): # Create the path path = c.add( k[f"{suffix} path"], "make_output_path", "config", name=f"{k.base.name}.{suffix}", ) # Write `key` to the path file_keys.append(c.add(k[suffix], "write_report", base_key, path)) # Write all files c.add(k["file"], file_keys) # Store data on "scenario" c.add(k["store"], "store_ts", "scenario", base_key) # Both write and store return single_key(c.add(k["all"], [k["file"], k["store"]]))
# Units for final energy. This *exact* value (and not e.g. "EJ / year") is required for # the legacy reporting to properly handle the result. fe_unit = "EJ/yr" CONVERT_IAMC = ( # NB these are currently tailored to produce the variable names expected for the # NGFS project dict( variable="transport activity", base="out:nl-t-ya-c:transport+units", var=["Energy Service|Transportation", "t", "c"], sums=["c", "t", "c-t"], ), dict( variable="transport stock", base="CAP:nl-t-ya:ldv+units", var=["Transport|Stock|Road|Passenger|LDV", "t"], unit="Mvehicle", ), dict( variable="transport sales", base="CAP_NEW:nl-t-yv:ldv+units", var=["Transport|Sales|Road|Passenger|LDV", "t"], unit="Mvehicle", ), # Final energy # # The following are 4 different partial sums of in::transport, in which # individual technologies are already aggregated to modes dict( variable="transport fe", base="in:nl-t-ya-c:transport+units", var=["Final Energy|Transportation", "t", "c"], sums=["c", "t", "c-t"], unit=fe_unit, ), dict( variable="transport fe ldv", base="in:nl-t-ya-c:ldv+units", var=["Final Energy|Transportation|Road|Passenger|LDV", "t", "c"], unit="EJ/yr", ), # Emissions using MESSAGEix emission_factor parameter # base: auto-sum over dimensions yv, m, h # var: Same as in data/report/global.yaml # dict( # variable="transport emi 0", # base="emi:nl-t-ya-e-gwp metric-e equivalent:gwpe+agg", # var=[ # "Emissions|CO2|Energy|Demand|Transportation", # "t", # "e", # "e equivalent", # "gwp metric", # ], # ), # dict( # variable="transport emi 1", # base="emi:nl-t-ya-e:transport+units", # var=["Emissions", "e", "Energy|Demand|Transportation", "t"], # sums=["t"], # unit="Mt/yr", # ), # # # For debugging # dict(variable="debug ACT", base="ACT:nl-t-ya", var=["DEBUG", "t"], unit="-"), # dict(variable="debug CAP", base="CAP:nl-t-ya", var=["DEBUG", "t"], unit="-"), # dict( # variable="debug CAP_NEW", base="CAP_NEW:nl-t-yv", var=["DEBUG", "t"], unit="-" # ), )
[docs]def convert_iamc(c: "Computer") -> "Key": """Add tasks from :data:`.CONVERT_IAMC`.""" from message_ix_models.report import iamc as handle_iamc from message_ix_models.report import util util.REPLACE_VARS.update({r"^CAP\|(Transport)": r"\1"}) keys = [] for info in CONVERT_IAMC: handle_iamc(c, deepcopy(info)) keys.append(f"{info['variable']}::iamc") # Concatenate IAMC-format tables c.add("transport::iamc", "concat", *keys) # Add tasks for writing IAMC-structured data to file and storing on the scenario return single_key(c.apply(add_iamc_store_write, "transport::iamc"))
[docs]def misc(c: "Computer") -> None: """Add miscellaneous tasks. Among others, these include: - ``calibrate fe`` → a file :file:`calibrate-fe.csv`. See the header comment. """ config: "Config" = c.graph["config"]["transport"] # Configuration for :func:`check`. Adds a single key, 'transport check', that # depends on others and returns a :class:`pandas.Series` of :class:`bool`. c.add("transport check", "transport_check", "scenario", "ACT:nl-t-yv-va-m-h") # Exogenous data c.add("distance:nl:non-ldv", "distance_nonldv", "config") # Demand per capita c.add("demand::capita", "div" "demand:n-c-y", "population:n-y") # Adjustment factor for LDV calibration: fuel economy ratio k_num = Key("in:nl-t-ya-c:transport+units") / "c" # As in CONVERT_IAMC k_denom = Key("out:nl-t-ya-c:transport+units") / "c" # As in CONVERT_IAMC k_check = single_key(c.add("fuel economy::check", "div", k_num, k_denom)) c.add( k_check + "sel", "select", k_check, indexers=dict(t="LDV", ya=config.base_model_info.y0), drop=True, ) k_ratio = single_key( c.add( "fuel economy::ratio", "div", "fuel economy:nl-m:ldv+ref", k_check + "sel" ) ) c.add("calibrate fe path", "make_output_path", "config", name="calibrate-fe.csv") hc = "\n\n".join( [ "Calibration factor for LDV fuel economy", f"Ratio of ldv-fuel-economy-ref.csv\n to ({k_num} / {k_denom})", "Units: dimensionless\n", ] ) c.add( "calibrate fe", "write_report", k_ratio, "calibrate fe path", kwargs=dict(header_comment=hc), )
[docs]def callback(rep: Reporter, context: Context) -> None: """:meth:`.prepare_reporter` callback for MESSAGEix-Transport. Among others, adds: - ``{in,out}::transport``: with outputs aggregated by technology group or "mode". - ``transport plots``: the plots from :mod:`.transport.plot`. If the scenario to be reported is not solved, only a subset of plots are added. - ``transport all``: all of the above. """ from . import base, build N_keys = len(rep.graph) # - Configure MESSAGEix-Transport. # - Add structure and other information. # - Call, inter alia: # - demand.prepare_computer() for ex-post mode and demand calculations # - plot.prepare_computer() for plots check = build.get_computer( context, obj=rep, visualize=False, scenario=rep.graph.get("scenario") ) assert check is rep # Same `rep` was returned if False: log.info("Filter out non-transport technologies") # Plain "transport" from the base model, for e.g. prices t_filter = {"transport"} # MESSAGEix-Transport -specific technologies t_filter.update(map(str, rep.get("t::transport").copy())) # # Required commodities (e.g. fuel) from the base model # t_filter.update(spec.require.set["commodity"]) rep.set_filters(t=sorted(t_filter)) # Configure replacements for conversion to IAMC data structure add_replacements("t", context.transport.spec.add.set["technology"]) # Apply some functions that prepare further tasks. Order matters here. aggregate(rep) select_transport_techs(rep) reapply_units(rep) misc(rep) iamc_key = convert_iamc(rep) # Add tasks that prepare data to parametrize the MESSAGEix-GLOBIOM base model base_key = base.prepare_reporter(rep) rep.add( "transport all", [ # Use ths line to both store and write to file IAMC structured-data iamc_key, # Use this line for "transport::iamc+file" instead of "transport::iamc+all" # iamc_key - "all" + "file", "transport plots", base_key, ], ) log.info(f"Added {len(rep.graph)-N_keys} keys")
# TODO Write an SVG visualization of reporting calculations
[docs]def configure_legacy_reporting(config: dict) -> None: """Callback to configure the legacy reporting.""" from message_ix_models.report.legacy.default_tables import COMMODITY # NB the legacy reporting doesn't pass a context object to the hook that calls this # function, so get an instance directly context = Context.get_instance() # If it does not already exist, read transport configuration onto the Context, # including reporting config context.setdefault("transport", Config.from_context(context)) # Get a spec spec: "Spec" = context.transport.spec # Commented: pp_utils._retr_act_data() raises IndexError if lists are empty # # Clear existing entries # # NB it should not have any effect to leave these in # for key in config: # if key.startswith("trp "): # # log.debug(f"Discard '{key}': {config[key]}") # config[key] = [] # Iterate over technologies in the transport model spec for t in spec.add.set["technology"]: try: # Retrieve the input commodity for this technology commodity = t.eval_annotation("input")["commodity"] except (TypeError, KeyError): # No annotation, or no "commodity" info commodity = None else: # Map to the shorthands used in legacy reporting commodity = COMMODITY.get(commodity) if commodity is None: # log.debug(f"{t}: No legacy reporting") # Verbose continue group = f"trp {commodity}" # log.debug(f"{t} → '{group}'") config[group].append(t.id)
[docs]def latest_reporting_from_file( info: ScenarioInfo, base_dir: Path ) -> Tuple[Any, int, pd.DataFrame]: """Locate and retrieve the latest reported output for the scenario `info`. The file :file:`transport.csv` is sought in a subdirectory of `base_dir` identified by :attr:`.ScenarioInfo.path`. .. todo:: Move upstream, to :mod:`message_ix_models`. Returns ------- tuple 1. The path of the file read. 2. :class:`int`: The scenario version corresponding to the data read. 3. :class:`pandas.DataFrame`: the data. If no data is found, all the elements are :any:`None`. """ dirs = sorted(base_dir.glob(info.path.replace("vNone", "v*")), reverse=True) for _dir in dirs: path = _dir.joinpath("transport.csv") if not path.exists(): log.info(f"Skip {_dir}; no file 'transport.csv'") continue path_version = int(path.parent.name.split("v")[-1]) return ( path, path_version, pd.read_csv(path).assign( Scenario=lambda df: df.Scenario + f"#{path_version}" ), ) return None, -1, pd.DataFrame()
[docs]def latest_reporting_from_platform( info: ScenarioInfo, platform: "ixmp.Platform", minimum_version: int = -1 ) -> Tuple[Any, int, pd.DataFrame]: """Retrieve the latest reported output for the scenario described by `info`. The time series data attached to a scenario on `platform` is retrieved. .. todo:: Move upstream, to :mod:`message_ix_models`. Returns ------- tuple 1. The :class:`.Scenario` object. 2. :class:`int`: The scenario version corresponding to the data read. 3. :class:`pandas.DataFrame`: the data. If no data is found or the latest version with reporting time series data is <= `minimum_version`, all the elements are :any:`None`. """ from message_ix import Scenario for _, row in ( platform.scenario_list(model=info.model, scen=info.scenario, default=False) .sort_values(["version"], ascending=False) .iterrows() ): if row.version <= minimum_version: log.info(f"{row.version} ≤ minimum {minimum_version}") break elif row.is_locked: log.info(f"Skip {info.url} {row.version}; locked") continue s = Scenario( platform, model=info.model, scenario=info.scenario, version=row.version ) if s.has_solution(): return ( s, row.version, s.timeseries().assign( # Scenario=lambda df: df.Scenario + f"v{row.version}" ), ) else: log.info(f"Skip {info.url} {row.version}; no reporting output") del s return None, -1, pd.DataFrame()
[docs]def multi(context: Context, targets): """Report outputs from multiple scenarios.""" import plotnine as p9 report_dir = context.get_local_path("report") platform = context.get_platform() dfs = [] for target in map(ScenarioInfo.from_url, targets): path, path_version, df_path = latest_reporting_from_file(target, report_dir) scen, scen_version, df_scen = latest_reporting_from_platform(target, platform) if path_version == scen_version == -1: raise RuntimeError(f"No reporting output available for {target}") elif path_version >= scen_version: source = "file" df = df_path version = path_version else: source = "platform" df = df_scen version = scen_version log.info(f"{target.url = } {source = } {version = }") dfs.append(df) # FIXME This duplicates code from message_ix_models.tools.exo_data; deduplicate def drop_unique(df, names) -> pd.DataFrame: names_list = names.split() for name in names_list: values = df[name].unique() if len(values) > 1: raise RuntimeError(f"Not unique {name!r}: {values}") return df.drop(names_list, axis=1) # Convert to a genno.Quantity cols = ["Variable", "Model", "Scenario", "Region", "Unit"] data = genno.Quantity( pd.concat(dfs) .sort_values(cols) .melt(id_vars=cols, var_name="y") .astype({"y": int}) .pipe(drop_unique, "Model") .rename(columns={"Variable": "v", "Scenario": "s", "Region": "n"}) .dropna(subset=["value"]) .set_index("v s n y Unit".split())["value"] ) # Select a subset of data qty = quantity_from_iamc(data, r"Transport\|Stock\|Road\|Passenger\|LDV\|(.*)") # Plot # TODO Move to .transport.plot plot = ( p9.ggplot(qty.to_dataframe().reset_index()) + p9.aes(x="y", y="value", color="v") + p9.facet_grid("n ~ s") + p9.geom_point() + p9.geom_line() + p9.theme(figure_size=(11.7, 16.6)) ) plot.save("debug.pdf") return data
[docs]def quantity_from_iamc(qty: "AnyQuantity", variable: str) -> "AnyQuantity": """Extract data for a single measure from `qty` with (at least) dimensions v, u. .. todo:: Move upstream, to either :mod:`ixmp` or :mod:`genno`. Parameters ---------- variable : str Regular expression to match the ``v`` dimension of `qty`. """ import re from genno.operator import relabel, select expr = re.compile(variable) variables, replacements = [], {} for var in qty.coords["v"].data: if match := expr.fullmatch(var): variables.append(match.group(0)) replacements[match.group(0)] = match.group(1) subset = qty.pipe(select, {"v": variables}).pipe(relabel, {"v": replacements}) unique_units = subset.coords["Unit"].data assert 1 == len(unique_units) subset.units = unique_units[0] return subset.sel(Unit=unique_units[0], drop=True)