from __future__ import annotations from typing import Optional import numpy as np from .config import ( DEMOGRAPHY_TYPE_IDS, GENERATOR_RECIPE_IDS, Config, ) from .demography import summarize_demography from .utils import bin_average_log10_ne, time_grid class StdpopsimSimulationError(RuntimeError): """Raised when a requested stdpopsim anchor cannot be generated.""" def _extract_single_population_trajectory(dem) -> tuple[list[float], list[float], list[float]]: """Extract first-population size and growth-rate epochs from msprime. This remains a proxy for structured models, but it exactly preserves native exponential growth for genuinely single-population stdpopsim models. """ first_pop = dem.populations[0] pop_name = first_pop.name pop_id = getattr(first_pop, "id", 0) current_size = float(first_pop.initial_size) current_rate = float(first_pop.growth_rate) previous_time = 0.0 breaks: list[float] = [] values = [current_size] growth_rates = [current_rate] events = sorted(getattr(dem, "events", []), key=lambda event: getattr(event, "time", 0)) for event in events: if event.__class__.__name__ != "PopulationParametersChange": continue population = getattr(event, "population", None) population_id = getattr(event, "population_id", None) if population is not None: applies = population in {-1, 0, pop_id, pop_name} elif population_id is not None: applies = population_id in {-1, 0, pop_id} else: applies = True if not applies: continue event_time = float(event.time) current_size *= float(np.exp(-current_rate * (event_time - previous_time))) if getattr(event, "initial_size", None) is not None: current_size = float(event.initial_size) if getattr(event, "growth_rate", None) is not None: current_rate = float(event.growth_rate) if event_time <= 0: values[0] = current_size growth_rates[0] = current_rate elif breaks and np.isclose(event_time, breaks[-1], rtol=0.0, atol=1e-12): values[-1] = current_size growth_rates[-1] = current_rate else: breaks.append(event_time) values.append(current_size) growth_rates.append(current_rate) previous_time = event_time return breaks, values, growth_rates def validate_stdpopsim_setup(cfg: Config) -> None: """Fail before generation if an explicitly enabled anchor setup is unusable.""" if not cfg.enable_stdpopsim or cfg.p_stdpopsim_anchor <= 0.0: return try: import stdpopsim # type: ignore species = stdpopsim.get_species(cfg.stdpopsim_species) model_ids = [model.strip() for model in cfg.stdpopsim_models.split(",") if model.strip()] if model_ids: for model_id in model_ids: species.get_demographic_model(model_id) elif not getattr(species, "demographic_models", None): raise ValueError(f"species {cfg.stdpopsim_species!r} has no demographic models") stdpopsim.get_engine("msprime") except Exception as exc: raise StdpopsimSimulationError( "stdpopsim anchors are enabled but the dependency, species, models, " f"or engine are unavailable: {exc}" ) from exc def try_stdpopsim_sample( rng: np.random.Generator, cfg: Config, ) -> Optional[tuple[object, np.ndarray, dict, object]]: if not cfg.enable_stdpopsim: return None try: import stdpopsim # type: ignore species = stdpopsim.get_species(cfg.stdpopsim_species) model_ids = [m.strip() for m in cfg.stdpopsim_models.split(",") if m.strip()] if not model_ids: model_ids = [m.id for m in species.demographic_models[: min(5, len(species.demographic_models))]] model_id = str(rng.choice(model_ids)) model = species.get_demographic_model(model_id) contig_length = cfg.stdpopsim_contig_length if cfg.stdpopsim_contig_length > 0 else cfg.seq_len contig = species.get_contig( length=contig_length, mutation_rate=cfg.baseline_mu, recombination_rate=cfg.baseline_rec, ) pop_id = model.populations[0].id if getattr(model, "populations", None) else "pop_0" n_dip = max(1, cfg.n_haplotypes // 2) try: samples = model.get_samples(n_dip, *([0] * max(0, len(model.populations) - 1))) except Exception: samples = {pop_id: n_dip} engine = stdpopsim.get_engine("msprime") ts = engine.simulate(model, contig, samples, seed=int(rng.integers(1, 2**31 - 2))) try: ts = ts.trim() except Exception: pass edges, _ = time_grid(cfg) breaks: list[float] = [] values: list[float] = [] growth_rates: list[float] = [] dem = getattr(model, "model", None) if dem is None: raise ValueError( f"stdpopsim model {model_id!r} exposes no msprime demography" ) breaks, values, growth_rates = _extract_single_population_trajectory(dem) if not values: raise ValueError( f"stdpopsim model {model_id!r} has no extractable population trajectory" ) if not growth_rates: raise ValueError( f"stdpopsim model {model_id!r} has no extractable growth-rate trajectory" ) y = bin_average_log10_ne(edges, breaks, values, growth_rates) trajectory_meta = { "source_type": "stdpopsim_anchor", "target_quality": "heuristic_single_population", "native_continuous_growth": any( float(rate) != 0.0 for rate in growth_rates ), } meta = { **trajectory_meta, **summarize_demography( edges, breaks, values, growth_rates, trajectory_meta, generator_recipe="stdpopsim", demography_type="stdpopsim", ), "scenario": "stdpopsim", "generator_recipe": "stdpopsim", "generator_recipe_id": GENERATOR_RECIPE_IDS["stdpopsim"], "demography_type": "stdpopsim", "demography_type_id": DEMOGRAPHY_TYPE_IDS["stdpopsim"], "demography_mixture_version": cfg.demography_mixture_version, "stdpopsim_species": cfg.stdpopsim_species, "stdpopsim_model_id": model_id, "stdpopsim_population": pop_id, "stdpopsim_contig_length": int(contig_length), "target_scale": "log10_Ne", "target_aggregation": "log_time_bin_average", "target_time_bins": int(cfg.time_bins), "demography_resolution_log10": float(cfg.demography_resolution_log10), "demography_breaks": [float(x) for x in breaks], "demography_values": [float(x) for x in values], "demography_growth_rates": [float(x) for x in growth_rates], "demography_function": ( "piecewise_exponential_linear_time" if any(float(rate) != 0.0 for rate in growth_rates) else "piecewise_constant" ), } return ts, y, meta, contig except Exception as exc: raise StdpopsimSimulationError(f"failed to generate a stdpopsim anchor: {exc}") from exc