Source code for odatse.algorithm.pamc

# SPDX-License-Identifier: MPL-2.0
#
# ODAT-SE -- an open framework for data analysis
# Copyright (C) 2020- The University of Tokyo
#
# This Source Code Form is subject to the terms of the Mozilla Public License, v. 2.0.
# If a copy of the MPL was not distributed with this file, You can obtain one at http://mozilla.org/MPL/2.0/.

from typing import Union, Optional, TYPE_CHECKING

from io import open
import copy
import time

import numpy as np

import odatse
import odatse.exception
import odatse.algorithm.montecarlo
from odatse.algorithm.gather import gather_replica, gather_data
import odatse.util.resampling
from odatse.util.read_ts import read_Ts
from odatse.util.data_writer import DataWriter

if TYPE_CHECKING:
    from mpi4py import MPI


[docs] class Algorithm(odatse.algorithm.montecarlo.AlgorithmBase): """ Population Annealing Monte Carlo (PAMC) Algorithm Implementation. PAMC is an advanced Monte Carlo method that combines features of simulated annealing and population-based methods. It maintains a population of walkers that evolves through a sequence of temperatures, using both Monte Carlo updates and resampling. Key Features: - Maintains a population of walkers that evolves through temperature steps - Uses importance sampling with weight-based resampling - Supports both fixed and variable population sizes - Calculates free energy differences between temperature steps - Provides error estimates through population statistics Algorithm Flow: 1. Initialize walker population at highest temperature 2. For each temperature step: a. Perform Monte Carlo updates at current temperature b. Calculate weights for next temperature c. Resample population based on weights d. Update statistical estimates 3. Track best solutions and maintain system statistics Attributes ---------- logweights : np.ndarray Log of importance weights for each walker fx : np.ndarray Current energy/objective values nreplicas : np.ndarray Number of replicas at each temperature betas : np.ndarray Inverse temperatures (β = 1/T) Tindex : int Current temperature index Fmeans : np.ndarray Mean free energy at each temperature Ferrs : np.ndarray Free energy error estimates walker_ancestors : np.ndarray Tracks genealogy of walkers for analysis """ # Coordinate bounds/steps live on self.statespace; the walker state is in # self.state (see montecarlo.AlgorithmBase / state.py). numsteps: int numsteps_annealing: int logweights: np.ndarray fx: np.ndarray istep: int nreplicas: np.ndarray betas: np.ndarray Tindex: int fx_from_reset: np.ndarray # 0th column: number of accepted MC trials # 1st column: number of total MC trials naccepted_from_reset: np.ndarray resampling_interval: int nset_bootstrap: int walker_ancestors: np.ndarray populations: np.ndarray family_lo: int family_hi: int Fmeans: np.ndarray Ferrs: np.ndarray # PAMC-specific fields appended to the MC base checkpoint. # Fields restored with slice assignment or special logic are still listed # here so that ``__getstate__`` saves them; ``_apply_state`` handles them # explicitly rather than via the generic setattr loop. _checkpoint_attrs: list[str] = [ "betas", "nwalkers", "input_as_beta", "numsteps_for_T", "Tindex", "index_from_reset", "logZ", "logZs", "logweights", "Fmeans", "Ferrs", "nreplicas", "populations", "family_lo", "family_hi", "walker_ancestors", "fx_from_reset", "naccepted_from_reset", "acceptance_ratio", "pr_list", ]
[docs] def __init__( self, info: odatse.Info, runner: odatse.Runner = None, run_mode: str = "initial", ) -> None: """ Initialize the Algorithm class. Parameters ---------- info : odatse.Info Information object containing algorithm parameters. runner : odatse.Runner, optional Runner object for executing the algorithm, by default None. run_mode : str, optional Mode in which to run the algorithm, by default "initial". """ info_pamc = info.algorithm["pamc"] nwalkers = info_pamc.get("nreplica_per_proc", 1) super().__init__(info=info, runner=runner, nwalkers=nwalkers, run_mode=run_mode) self.verbose = True and odatse.mpi.algrank() is not None and odatse.mpi.algrank() == 0 numT = self._find_scheduling(info_pamc) self.input_as_beta, self.betas = read_Ts(info_pamc, numT=numT) self.betas.sort() self.fix_nwalkers = info_pamc.get("fix_num_replicas", True) self.resampling_interval = info_pamc.get("resampling_interval", 1) if self.resampling_interval < 1: self.resampling_interval = numT + 1 self.export_combined_files = info_pamc.get("export_combined_files", False) self.separate_T = info_pamc.get("separate_T", True) self.anneal_from_beta0 = self.betas[0] > 0.0 and info_pamc.get( "anneal_from_beta0", False )
[docs] def _initialize(self) -> None: super()._initialize() # PAMC-specific counters for a fresh run self.Tindex = 0 self.index_from_reset = 0 self.istep = 0 numT = len(self.betas) self.logZ = 0.0 self.logZs = np.zeros(numT) self.logweights = np.zeros(self.nwalkers) self.Fmeans = np.zeros(numT) self.Ferrs = np.zeros(numT) nreplicas = odatse.mpi.algsize() * self.nwalkers self.nreplicas = np.full(numT, nreplicas) self.populations = np.zeros((numT, self.nwalkers), dtype=int) self.family_lo = self.nwalkers * odatse.mpi.algrank() self.family_hi = self.nwalkers * (odatse.mpi.algrank() + 1) self.walker_ancestors = np.arange(self.family_lo, self.family_hi) self.fx_from_reset = np.zeros((self.resampling_interval, self.nwalkers)) self.naccepted_from_reset = np.zeros((self.resampling_interval, 2), dtype=int) self.acceptance_ratio = np.zeros(numT) self.pr_list = np.zeros(numT) self._show_parameters()
[docs] def _find_scheduling(self, info_pamc) -> int: """ Determine the temperature schedule and number of steps. The schedule can be specified in three ways: 1. Total steps and steps per temperature 2. Total steps and number of temperatures 3. Steps per temperature and number of temperatures The method ensures even distribution of computational effort across temperature steps while respecting the specified constraints. Parameters ---------- info_pamc : dict Configuration dictionary containing: - numsteps: Total number of Monte Carlo steps - numsteps_annealing: Steps per temperature - Tnum: Number of temperature points Returns ------- int Number of temperature steps in the schedule Raises ------ odatse.exception.InputError If the scheduling parameters are inconsistent """ numsteps = info_pamc.get("numsteps", 0) numsteps_annealing = info_pamc.get("numsteps_annealing", 0) numT = info_pamc.get("Tnum", 0) oks = np.array([numsteps, numsteps_annealing, numT]) > 0 if np.count_nonzero(oks) != 2: msg = "Two of 'numsteps', 'numsteps_annealing', " msg += "and 'Tnum' should be positive in the input file\n" msg += f" numsteps = {numsteps}\n" msg += f" numsteps_annealing = {numsteps_annealing}\n" msg += f" Tnum = {numT}\n" raise odatse.exception.InputError(msg) if numsteps <= 0: self.numsteps_for_T = np.full(numT, numsteps_annealing) elif numsteps_annealing <= 0: nr = numsteps // numT self.numsteps_for_T = np.full(numT, nr) rem = numsteps - nr * numT if rem > 0: # Distribute the remainder over the first ``rem`` temperatures # so that the per-temperature step counts sum to ``numsteps``. self.numsteps_for_T[0:rem] += 1 else: ss: list[int] = [] while numsteps > 0: if numsteps > numsteps_annealing: ss.append(numsteps_annealing) numsteps -= numsteps_annealing else: ss.append(numsteps) numsteps = 0 self.numsteps_for_T = np.array(ss) numT = len(ss) return numT
[docs] def _run(self) -> None: # dispatch は prepare() が処理済み writer = self._setup_writer() if self.mode.startswith("init"): beta = self.betas[self.Tindex] self.fx = self._evaluate(self.state) if self.anneal_from_beta0: # Anneal from beta=0 and resample # In _resample, self.logZ is updated self.logweights = -self.betas[0] * self.fx self._resample(at_init=True) self._write_result(writer["trial"], [np.exp(self.logweights), self.walker_ancestors]) self._write_result(writer["result"], [np.exp(self.logweights), self.walker_ancestors]) self.istep += 1 minidx = np.argmin(self.fx) self.best_x = copy.copy(self.state.x[minidx, :]) self.best_fx = np.min(self.fx[minidx]) self.best_istep = 0 self.best_iwalker = 0 next_checkpoint_step = self.istep + self.checkpoint_steps next_checkpoint_time = time.time() + self.checkpoint_interval if self.verbose: print("\u03b2 mean[f] Err[f] nreplica log(Z/Z0) acceptance_ratio") numT = len(self.betas) while self.Tindex < numT: # print(">>> Tindex = {}".format(self.Tindex)) Tindex = self.Tindex beta = self.betas[self.Tindex] if self.nwalkers != 0: for _ in range(self.numsteps_for_T[Tindex]): self.local_update( beta, extra_info_to_write=[ np.exp(self.logweights), self.walker_ancestors, ], ) self.istep += 1 else : #self.nwalkers == 0 pass # print(">>> istep={}".format(self.istep)) self.fx_from_reset[self.index_from_reset, :] = self.fx[:] self.naccepted_from_reset[self.index_from_reset, 0] = self.naccepted self.naccepted_from_reset[self.index_from_reset, 1] = self.ntrial self.naccepted = 0 self.ntrial = 0 self.index_from_reset += 1 # write weights if writer["weight"]: for iwalker in range(self.nwalkers): writer["weight"].write(Tindex, beta, iwalker, 0, self.fx[iwalker], self.logweights[iwalker], *self.state.x[iwalker,:]) # calculate participation ratio pr = self._calc_participation_ratio() self.pr_list[Tindex] = pr if Tindex == numT - 1: break dbeta = self.betas[Tindex + 1] - self.betas[Tindex] self.logweights += -dbeta * self.fx if self.index_from_reset == self.resampling_interval: time_sta = time.perf_counter() self._resample() time_end = time.perf_counter() self.timer["run"]["resampling"] += time_end - time_sta self.index_from_reset = 0 self.Tindex += 1 if self.checkpoint: time_now = time.time() if self.istep >= next_checkpoint_step or time_now >= next_checkpoint_time: print(">>> checkpoint") self._save_state(self.checkpoint_file) next_checkpoint_step = self.istep + self.checkpoint_steps next_checkpoint_time = time_now + self.checkpoint_interval # store final state for continuation if self.checkpoint: print(">>> store final state") self._save_state(self.checkpoint_file) if self.index_from_reset > 0: res = self._gather_information(self.index_from_reset) self._save_stats(res) # must close explicitly self._close_writer(writer) if self.separate_T and not self.export_combined_files: self._split_result_file("trial") self._split_result_file("result") if odatse.mpi.algsize() > 1: odatse.mpi.algcomm().barrier() print("complete main process : rank {:08d}/{:08d}".format(odatse.mpi.algrank(), odatse.mpi.algsize()))
def _setup_writer(self): write_mode = "w" if self.mode.startswith("init") else "a" item_list = [ "step", "walker", ("beta" if self.input_as_beta else "T"), "fx", *self.label_list, "weight", "ancestor", ] file_trial = DataWriter("trial.txt", mode=write_mode, item_list=item_list, combined=self.export_combined_files) file_result = DataWriter("result.txt", mode=write_mode, item_list=item_list, combined=self.export_combined_files) self._set_writer(file_trial, file_result) item_list_weight = [ "Tindex", "beta", "walker", "idnum", "fx", "log_weight", *self.label_list, ] file_weight = DataWriter("weight.txt", mode=write_mode, item_list=item_list_weight, combined=self.export_combined_files) return { "trial": file_trial, "result": file_result, "weight": file_weight } def _close_writer(self, writers): for k, v in writers.items(): if v: v.close()
[docs] def _gather_information(self, numT: int = None) -> dict[str, np.ndarray]: """ Collect and organize statistical information across all processes. Gathers data needed for: 1. Free energy calculations 2. Error estimation 3. Population statistics 4. Acceptance rate monitoring Parameters ---------- numT : int, optional Number of temperature steps to gather data for Returns ------- dict[str, np.ndarray] Contains: - fxs: Energy values for all walkers - logweights: Log of importance weights - ns: Number of walkers per process - ancestors: Genealogical tracking data - acceptance ratio: MC acceptance rates """ if numT is None: numT = self.resampling_interval res = {} res["ns"] = gather_data(np.array([self.nwalkers])) res["fxs"] = gather_replica(self.fx_from_reset[0:numT,:], axis=1) res["ancestors"] = gather_replica(self.walker_ancestors) nacc = gather_data([self.naccepted_from_reset[0:numT,:]]) nacc = np.sum(nacc, axis=0) # A temperature with no trials (e.g. numsteps_for_T == 0) would give a # 0/0 acceptance ratio; report 0.0 there instead of nan/inf. ntrials = nacc[:, 1] res["acceptance ratio"] = np.divide( nacc[:, 0], ntrials, out=np.zeros(nacc.shape[0], dtype=np.float64), where=(ntrials != 0), ) fxs = res["fxs"] nreplicas = np.sum(res["ns"]) betas = self.betas[self.Tindex-numT+1 : self.Tindex+1] lw = np.zeros((numT, nreplicas), dtype=np.float64) for i in range(1, numT): dbeta = betas[i] - betas[i-1] lw[i, :] = lw[i-1, :] - dbeta * fxs[i-1, :] res["logweights"] = lw return res
[docs] def _save_stats(self, info: dict[str, np.ndarray]) -> None: """ Calculate and save statistical measures from the simulation. Performs: 1. Free energy calculations using weighted averages 2. Error estimation using jackknife resampling 3. Population size tracking 4. Acceptance rate monitoring 5. Partition function estimation Uses bias-corrected jackknife for reliable error estimates of weighted averages in the presence of correlations. Parameters ---------- info : dict[str, np.ndarray] Dictionary containing the following keys: - fxs: Objective function of each walker over all processes. - logweights: Logarithm of weights. - ns: Number of walkers in each process. - ancestors: Ancestor (origin) of each walker. - acceptance ratio: Acceptance ratio for each temperature. """ fxs = info["fxs"] numT, nreplicas = fxs.shape endTindex = self.Tindex + 1 startTindex = endTindex - numT logweights = info["logweights"] logweights_max = logweights.max(axis=1).reshape(-1, 1) weights = np.exp(logweights - logweights_max) # to avoid overflow # Bias-corrected jackknife resampling method fs = np.zeros((numT, nreplicas)) fw_sum = (fxs * weights).sum(axis=1) w_sum = weights.sum(axis=1) for i in range(nreplicas): F = fw_sum - fxs[:, i] * weights[:, i] W = w_sum - weights[:, i] fs[:, i] = F / W N = fs.shape[1] fm = N * (fw_sum / w_sum) - (N - 1) * fs.mean(axis=1) ferr = np.sqrt((N - 1) * fs.var(axis=1)) self.Fmeans[startTindex:endTindex] = fm self.Ferrs[startTindex:endTindex] = ferr self.nreplicas[startTindex:endTindex] = nreplicas self.acceptance_ratio[startTindex:endTindex] = info["acceptance ratio"][0:numT] logz = np.log(np.mean(weights, axis=1)) logz += logweights_max.flatten() self.logZs[startTindex:endTindex] = self.logZ + logz if endTindex < len(self.betas): bdiff = self.betas[endTindex] - self.betas[endTindex - 1] w = np.exp(logweights[-1, :] - bdiff * fxs[-1, :]) self.logZ = self.logZs[startTindex] + np.log(w.mean()) if self.verbose: for iT in range(startTindex, endTindex): print(" ".join(map(str, [ self.betas[iT], self.Fmeans[iT], self.Ferrs[iT], self.nreplicas[iT], self.logZs[iT], self.acceptance_ratio[iT], ])))
[docs] def _resample(self, at_init: bool = False) -> None: """ Perform population resampling between temperature steps. This is a key component of PAMC that: 1. Gathers current population statistics (unless at_init) 2. Calculates importance weights for the temperature change 3. Resamples walkers based on their weights 4. Updates population statistics and free energy estimates (unless at_init) Parameters ---------- at_init : bool, optional If True, use current logweights only and skip gather/save_stats (used when resampling at init with anneal_from_beta0). self.logZ is updated in this case. The resampling can be done in two modes: - Fixed: Maintains constant population size - Varied: Allows population size to fluctuate based on weights Implementation Details: - Uses log-weights to prevent numerical overflow - Maintains walker genealogy for analysis - Updates free energy estimates using resampling data - Handles MPI communication for parallel execution """ if at_init: logweights = gather_replica(self.logweights) lw_max = logweights.max() weights = np.exp(logweights - lw_max) self.logZ = np.log(weights.mean()) + lw_max ns = gather_data(np.array([self.nwalkers])) else: res = self._gather_information() self._save_stats(res) dbeta = self.betas[self.Tindex + 1] - self.betas[self.Tindex] logweights = res["logweights"][-1, :] - dbeta * res["fxs"][-1, :] weights = np.exp(logweights - logweights.max()) ns = res["ns"] if self.fix_nwalkers: self._resample_fixed(weights) self.logweights[:] = 0.0 else: offsets = np.cumsum(ns) - ns self._resample_varied(weights, offsets[odatse.mpi.algrank()]) self.fx_from_reset = np.zeros((self.resampling_interval, self.nwalkers)) self.logweights = np.zeros(self.nwalkers)
[docs] def _resample_fixed(self, weights: np.ndarray) -> None: """ Perform resampling with fixed weights. This method resamples the walkers based on the provided weights and updates the state of the algorithm accordingly. Parameters ---------- weights : np.ndarray Array of weights for resampling. """ resampler = odatse.util.resampling.WalkerTable(weights) new_index = resampler.sample(self.rng, self.nwalkers) states = self.statespace.gather(self.state) ancestors = gather_replica(self.walker_ancestors) self.state = self.statespace.pick(states, new_index) self.walker_ancestors = ancestors[new_index] fxs = gather_replica(self.fx) self.fx = fxs[new_index]
[docs] def _resample_varied(self, weights: np.ndarray, offset: int) -> None: """ Resample population allowing size variation. Uses Poisson resampling: 1. Calculates expected number of copies from weights 2. Samples actual copies using Poisson distribution 3. Creates new population with variable size 4. Updates all walker properties Parameters ---------- weights : np.ndarray Importance weights for resampling offset : int Process-specific offset in global population """ weights_sum = np.sum(weights) expected_numbers = (self.nreplicas[0] / weights_sum) * weights[ offset : offset + self.nwalkers ] next_numbers = self.rng.poisson(expected_numbers) new_index = [] for iwalker, num in enumerate(next_numbers): new_index.extend([iwalker] * num) new_state = self.statespace.pick(self.state, new_index) new_fx = self.fx[new_index] self.state = new_state self.fx = new_fx self.walker_ancestors = np.array(new_index) # keep nwalkers a plain int (np.sum returns np.int64), since downstream # code does isinstance(..., int) checks and uses it as an array size self.nwalkers = int(np.sum(next_numbers))
[docs] def _calc_participation_ratio(self) -> float: """ Calculate the participation ratio of the current walker population. The participation ratio is a measure of the effective sample size and indicates how evenly distributed the weights are among walkers. It is calculated as (sum(w))²/sum(w²), where w are the normalized weights. A value close to the total number of walkers indicates well-distributed weights, while a small value indicates that only a few walkers dominate the population. To avoid numerical issues with large log-weights, we normalize by subtracting the maximum log-weight before exponentiation. Parameters ---------- None Uses the current state of self.logweights Returns ------- float The participation ratio, a value between 1 and the total number of walkers Notes ----- In parallel execution, this method aggregates weights across all processes to calculate the global participation ratio. """ log_weights = gather_replica(self.logweights) max_log_weight = np.max(log_weights) # Degenerate weights (e.g. all -inf) make (log_weights - max) contain # nan; the guard below turns that into a finite 0.0, so suppress the # transient invalid-value warning here. with np.errstate(invalid="ignore"): sum_weight = np.sum(np.exp(log_weights - max_log_weight)) sum_weight_sq = np.sum(np.exp(log_weights - max_log_weight)**2) # sum_weight_sq is normally >= 1 (the max element contributes exp(0)=1), # but degenerate weights make it 0 or nan; guard so the participation # ratio is a finite number rather than nan. pr = sum_weight ** 2 / sum_weight_sq if sum_weight_sq > 0 else 0.0 return pr
[docs] def _prepare(self) -> None: """ Prepare the algorithm for execution. Initialises the run-phase timers, then for ``continue`` mode applies the logweight update and advances ``Tindex`` to the next temperature. This must run after ``_apply_state()`` has extended the beta schedule (done inside ``_prepare()`` dispatch) and after the timers are ready. """ self.timer["run"]["submit"] = 0.0 self.timer["run"]["resampling"] = 0.0 if self.mode.startswith("continue"): Tindex = self.Tindex dbeta = self.betas[Tindex + 1] - self.betas[Tindex] self.logweights += -dbeta * self.fx if self.index_from_reset == self.resampling_interval: time_sta = time.perf_counter() self._resample() time_end = time.perf_counter() self.timer["run"]["resampling"] += time_end - time_sta self.index_from_reset = 0 self.Tindex += 1
def _split_result_file(self, tag): current_beta = -1 header_str = "" idx = 0 fwrite = None with open(tag + ".txt", "r") as fread: for line in fread: if line.startswith("#"): header_str += line else: data = line.split() beta = data[2] if beta != current_beta: if fwrite: fwrite.close() idx += 1 fwrite = open(tag + f"_T{idx}.txt", "w") fwrite.write(header_str) current_beta = beta fwrite.write(line) if fwrite: fwrite.close()
[docs] def _post(self) -> dict: """ Post-processing after the algorithm execution. Gathers the best solution across all processes and writes it to ``best_result.txt``, and writes the per-temperature statistics (free energy, errors, partition function) to ``fx.txt`` and the participation ratios to ``pr.txt``. (Splitting of ``trial``/``result`` into per-temperature files is done in ``_run``, not here.) """ best_fx = gather_data(np.array([self.best_fx])) best_x = gather_data(np.array([self.best_x])) best_istep = gather_data(np.array([self.best_istep])) best_iwalker = gather_data(np.array([self.best_iwalker])) best_rank = np.argmin(best_fx) if odatse.mpi.algrank() is not None and odatse.mpi.algrank() == 0: with open("best_result.txt", "w") as f: f.write(f"nprocs = {odatse.mpi.algsize()}\n") f.write(f"rank = {best_rank}\n") f.write(f"step = {best_istep[best_rank]}\n") f.write(f"walker = {best_iwalker[best_rank]}\n") f.write(f"fx = {best_fx[best_rank]}\n") for label, x in zip(self.label_list, best_x[best_rank]): f.write(f"{label} = {x}\n") print("Best Result:") print(f" rank = {best_rank}") print(f" step = {best_istep[best_rank]}") print(f" walker = {best_iwalker[best_rank]}") print(f" fx = {best_fx[best_rank]}") for label, x in zip(self.label_list, best_x[best_rank]): print(f" {label} = {x}") with open("fx.txt", "w") as f: f.write("# $1: 1/T\n") f.write("# $2: mean of f(x)\n") f.write("# $3: standard error of f(x)\n") f.write("# $4: number of replicas\n") f.write("# $5: log(Z/Z0)\n") f.write("# $6: acceptance ratio\n") for i in range(len(self.betas)): f.write(f"{self.betas[i]}") f.write(f" {self.Fmeans[i]} {self.Ferrs[i]}") f.write(f" {self.nreplicas[i]}") f.write(f" {self.logZs[i]}") f.write(f" {self.acceptance_ratio[i]}") f.write("\n") with open("pr.txt", "w") as f: f.write("# $1: Tindex\n") f.write("# $2: 1/T\n") f.write("# $3: participation ratio\n") for i in range(len(self.betas)): f.write(f"{i}") f.write(f" {self.betas[i]}") f.write(f" {self.pr_list[i]}") f.write("\n") return { "x": best_x[best_rank], "fx": best_fx[best_rank], "nprocs": odatse.mpi.algsize(), "rank": best_rank, "step": best_istep[best_rank], "walker": best_iwalker[best_rank], }
[docs] def _apply_state(self, data: dict, mode: str = "resume", restore_rng: bool = True) -> None: """Restore algorithm state from a checkpoint snapshot. Delegates MPI validation, RNG restore, and MC-layer fields to the base class, then handles PAMC-specific fields explicitly. Simple scalar fields are set directly; array fields that may be shorter than the current schedule (``logZs``, ``Fmeans``, etc.) are written with slice assignment after re-initialisation. The logweight update, optional resampling, and ``Tindex`` increment for ``continue`` mode are performed in ``prepare()`` after this method returns, so that the run-phase timers are already initialised. Parameters ---------- data : dict Snapshot previously produced by ``__getstate__``. mode : str ``"resume"`` — validate that the temperature schedule is identical to the one stored in *data*. ``"continue"`` — concatenate the stored schedule with the new one (the stored last beta must equal the new first beta). restore_rng : bool When *True* (default) the RNG state is restored from *data*; when *False* a fresh RNG state is kept (``--reset_rand`` mode). """ super()._apply_state(data, mode=mode, restore_rng=restore_rng) # -- simple scalar fields self.Tindex = data["Tindex"] self.index_from_reset = data["index_from_reset"] self.nwalkers = data["nwalkers"] # -- temperature schedule (mode-dependent) if mode == "resume": assert np.all(data["betas"] == self.betas) assert data["input_as_beta"] == self.input_as_beta assert np.all(data["numsteps_for_T"] == self.numsteps_for_T) assert self.Tindex < len(self.betas) elif mode == "continue": assert data["input_as_beta"] == self.input_as_beta if not data["betas"][-1] == self.betas[0]: raise odatse.exception.InputError( "temperature is not continuous between the saved and " "current schedules (saved last beta " f"{data['betas'][-1]} != current first beta {self.betas[0]})" ) self.betas = np.concatenate([data["betas"], self.betas[1:]]) self.numsteps_for_T = np.concatenate([data["numsteps_for_T"], self.numsteps_for_T[1:]]) # -- re-initialise length-numT arrays, then overwrite with saved data numT = len(self.betas) nreplicas = odatse.mpi.algsize() * self.nwalkers self.logZs = np.zeros(numT) self.Fmeans = np.zeros(numT) self.Ferrs = np.zeros(numT) self.nreplicas = np.full(numT, nreplicas) self.acceptance_ratio = np.zeros(numT) self.pr_list = np.zeros(numT) self.logZ = data["logZ"] self.logZs[:len(data["logZs"])] = data["logZs"] self.logweights = data["logweights"] self.Fmeans[:len(data["Fmeans"])] = data["Fmeans"] self.Ferrs[:len(data["Ferrs"])] = data["Ferrs"] self.nreplicas[:len(data["nreplicas"])] = data["nreplicas"] self.populations = data["populations"].copy() self.family_lo = data["family_lo"] self.family_hi = data["family_hi"] self.fx_from_reset = data["fx_from_reset"] self.walker_ancestors = data["walker_ancestors"] self.naccepted_from_reset = data["naccepted_from_reset"] self.acceptance_ratio[:len(data["acceptance_ratio"])] = data["acceptance_ratio"] self.pr_list[:len(data["pr_list"])] = data["pr_list"] self.statespace.rng = self.rng