freqtrade_origin/freqtrade/optimize/hyperopt/hyperopt.py

# pragma pylint: disable=too-many-instance-attributes, pointless-string-statement

"""
This module contains the hyperopt logic
"""

import logging
import random
import sys
from datetime import datetime
from math import ceil
from pathlib import Path
from typing import Any

import rapidjson
from joblib import Parallel, cpu_count, delayed, wrap_non_picklable_objects
from joblib.externals import cloudpickle
from rich.console import Console

from freqtrade.constants import FTHYPT_FILEVERSION, LAST_BT_RESULT_FN, Config
from freqtrade.enums import HyperoptState
from freqtrade.exceptions import OperationalException
from freqtrade.misc import file_dump_json, plural

# Import IHyperOpt and IHyperOptLoss to allow unpickling classes from these modules
from freqtrade.optimize.hyperopt.hyperopt_auto import HyperOptAuto
from freqtrade.optimize.hyperopt.hyperopt_optimizer import HyperOptimizer
from freqtrade.optimize.hyperopt.hyperopt_output import HyperoptOutput
from freqtrade.optimize.hyperopt_loss.hyperopt_loss_interface import IHyperOptLoss
from freqtrade.optimize.hyperopt_tools import (
    HyperoptStateContainer,
    HyperoptTools,
    hyperopt_serializer,
)
from freqtrade.util import get_progress_tracker


logger = logging.getLogger(__name__)


INITIAL_POINTS = 30

# Keep no more than SKOPT_MODEL_QUEUE_SIZE models
# in the skopt model queue, to optimize memory consumption
SKOPT_MODEL_QUEUE_SIZE = 10


class Hyperopt:
    """
    Hyperopt class, this class contains all the logic to run a hyperopt simulation

    To start a hyperopt run:
    hyperopt = Hyperopt(config)
    hyperopt.start()
    """

    def __init__(self, config: Config) -> None:
        self._hyper_out: HyperoptOutput = HyperoptOutput(streaming=True)

        self.config = config

        self.analyze_per_epoch = self.config.get("analyze_per_epoch", False)
        HyperoptStateContainer.set_state(HyperoptState.STARTUP)

        if self.config.get("hyperopt"):
            raise OperationalException(
                "Using separate Hyperopt files has been removed in 2021.9. Please convert "
                "your existing Hyperopt file to the new Hyperoptable strategy interface"
            )

        time_now = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
        strategy = str(self.config["strategy"])
        self.results_file: Path = (
            self.config["user_data_dir"]
            / "hyperopt_results"
            / f"strategy_{strategy}_{time_now}.fthypt"
        )
        self.data_pickle_file = (
            self.config["user_data_dir"] / "hyperopt_results" / "hyperopt_tickerdata.pkl"
        )
        self.total_epochs = config.get("epochs", 0)

        self.current_best_loss = 100

        self.clean_hyperopt()

        self.num_epochs_saved = 0
        self.current_best_epoch: dict[str, Any] | None = None

        if HyperoptTools.has_space(self.config, "sell"):
            # Make sure use_exit_signal is enabled
            self.config["use_exit_signal"] = True

        self.print_all = self.config.get("print_all", False)
        self.hyperopt_table_header = 0
        self.print_colorized = self.config.get("print_colorized", False)
        self.print_json = self.config.get("print_json", False)

        self.hyperopter = HyperOptimizer(self.config)

    @staticmethod
    def get_lock_filename(config: Config) -> str:
        return str(config["user_data_dir"] / "hyperopt.lock")

    def clean_hyperopt(self) -> None:
        """
        Remove hyperopt pickle files to restart hyperopt.
        """
        for f in [self.data_pickle_file, self.results_file]:
            p = Path(f)
            if p.is_file():
                logger.info(f"Removing `{p}`.")
                p.unlink()

    def hyperopt_pickle_magic(self, bases) -> None:
        """
        Hyperopt magic to allow strategy inheritance across files.
        For this to properly work, we need to register the module of the imported class
        to pickle as value.
        """
        for modules in bases:
            if modules.__name__ != "IStrategy":
                cloudpickle.register_pickle_by_value(sys.modules[modules.__module__])
                self.hyperopt_pickle_magic(modules.__bases__)

    def _save_result(self, epoch: dict) -> None:
        """
        Save hyperopt results to file
        Store one line per epoch.
        While not a valid json object - this allows appending easily.
        :param epoch: result dictionary for this epoch.
        """
        epoch[FTHYPT_FILEVERSION] = 2
        with self.results_file.open("a") as f:
            rapidjson.dump(
                epoch,
                f,
                default=hyperopt_serializer,
                number_mode=rapidjson.NM_NATIVE | rapidjson.NM_NAN,
            )
            f.write("\n")

        self.num_epochs_saved += 1
        logger.debug(
            f"{self.num_epochs_saved} {plural(self.num_epochs_saved, 'epoch')} "
            f"saved to '{self.results_file}'."
        )
        # Store hyperopt filename
        latest_filename = Path.joinpath(self.results_file.parent, LAST_BT_RESULT_FN)
        file_dump_json(latest_filename, {"latest_hyperopt": str(self.results_file.name)}, log=False)

    def print_results(self, results: dict[str, Any]) -> None:
        """
        Log results if it is better than any previous evaluation
        TODO: this should be moved to HyperoptTools too
        """
        is_best = results["is_best"]

        if self.print_all or is_best:
            self._hyper_out.add_data(
                self.config,
                [results],
                self.total_epochs,
                self.print_all,
            )

    def run_optimizer_parallel(self, parallel: Parallel, asked: list[list]) -> list[dict[str, Any]]:
        """Start optimizer in a parallel way"""
        return parallel(
            delayed(wrap_non_picklable_objects(self.hyperopter.generate_optimizer))(v)
            for v in asked
        )

    def _set_random_state(self, random_state: int | None) -> int:
        return random_state or random.randint(1, 2**16 - 1)  # noqa: S311

    def get_asked_points(self, n_points: int) -> tuple[list[list[Any]], list[bool]]:
        """
        Enforce points returned from `self.opt.ask` have not been already evaluated

        Steps:
        1. Try to get points using `self.opt.ask` first
        2. Discard the points that have already been evaluated
        3. Retry using `self.opt.ask` up to 3 times
        4. If still some points are missing in respect to `n_points`, random sample some points
        5. Repeat until at least `n_points` points in the `asked_non_tried` list
        6. Return a list with length truncated at `n_points`
        """

        def unique_list(a_list):
            new_list = []
            for item in a_list:
                if item not in new_list:
                    new_list.append(item)
            return new_list

        i = 0
        asked_non_tried: list[list[Any]] = []
        is_random_non_tried: list[bool] = []
        while i < 5 and len(asked_non_tried) < n_points:
            if i < 3:
                self.opt.cache_ = {}
                asked = unique_list(self.opt.ask(n_points=n_points * 5 if i > 0 else n_points))
                is_random = [False for _ in range(len(asked))]
            else:
                asked = unique_list(self.opt.space.rvs(n_samples=n_points * 5))
                is_random = [True for _ in range(len(asked))]
            is_random_non_tried += [
                rand
                for x, rand in zip(asked, is_random, strict=False)
                if x not in self.opt.Xi and x not in asked_non_tried
            ]
            asked_non_tried += [
                x for x in asked if x not in self.opt.Xi and x not in asked_non_tried
            ]
            i += 1

        if asked_non_tried:
            return (
                asked_non_tried[: min(len(asked_non_tried), n_points)],
                is_random_non_tried[: min(len(asked_non_tried), n_points)],
            )
        else:
            return self.opt.ask(n_points=n_points), [False for _ in range(n_points)]

    def evaluate_result(self, val: dict[str, Any], current: int, is_random: bool):
        """
        Evaluate results returned from generate_optimizer
        """
        val["current_epoch"] = current
        val["is_initial_point"] = current <= INITIAL_POINTS

        logger.debug("Optimizer epoch evaluated: %s", val)

        is_best = HyperoptTools.is_best_loss(val, self.current_best_loss)
        # This value is assigned here and not in the optimization method
        # to keep proper order in the list of results. That's because
        # evaluations can take different time. Here they are aligned in the
        # order they will be shown to the user.
        val["is_best"] = is_best
        val["is_random"] = is_random
        self.print_results(val)

        if is_best:
            self.current_best_loss = val["loss"]
            self.current_best_epoch = val

        self._save_result(val)

    def start(self) -> None:
        self.random_state = self._set_random_state(self.config.get("hyperopt_random_state"))
        logger.info(f"Using optimizer random state: {self.random_state}")
        self.hyperopt_table_header = -1
        self.hyperopter.prepare_hyperopt()

        cpus = cpu_count()
        logger.info(f"Found {cpus} CPU cores. Let's make them scream!")
        config_jobs = self.config.get("hyperopt_jobs", -1)
        logger.info(f"Number of parallel jobs set as: {config_jobs}")

        self.opt = self.hyperopter.get_optimizer(
            config_jobs, self.random_state, INITIAL_POINTS, SKOPT_MODEL_QUEUE_SIZE
        )

        try:
            with Parallel(n_jobs=config_jobs) as parallel:
                jobs = parallel._effective_n_jobs()
                logger.info(f"Effective number of parallel workers used: {jobs}")
                console = Console(
                    color_system="auto" if self.print_colorized else None,
                )

                # Define progressbar
                with get_progress_tracker(
                    console=console,
                    cust_callables=[self._hyper_out],
                ) as pbar:
                    task = pbar.add_task("Epochs", total=self.total_epochs)

                    start = 0

                    if self.analyze_per_epoch:
                        # First analysis not in parallel mode when using --analyze-per-epoch.
                        # This allows dataprovider to load it's informative cache.
                        asked, is_random = self.get_asked_points(n_points=1)
                        f_val0 = self.hyperopter.generate_optimizer(asked[0])
                        self.opt.tell(asked, [f_val0["loss"]])
                        self.evaluate_result(f_val0, 1, is_random[0])
                        pbar.update(task, advance=1)
                        start += 1

                    evals = ceil((self.total_epochs - start) / jobs)
                    for i in range(evals):
                        # Correct the number of epochs to be processed for the last
                        # iteration (should not exceed self.total_epochs in total)
                        n_rest = (i + 1) * jobs - (self.total_epochs - start)
                        current_jobs = jobs - n_rest if n_rest > 0 else jobs

                        asked, is_random = self.get_asked_points(n_points=current_jobs)
                        f_val = self.run_optimizer_parallel(parallel, asked)
                        self.opt.tell(asked, [v["loss"] for v in f_val])

                        for j, val in enumerate(f_val):
                            # Use human-friendly indexes here (starting from 1)
                            current = i * jobs + j + 1 + start

                            self.evaluate_result(val, current, is_random[j])
                            pbar.update(task, advance=1)

        except KeyboardInterrupt:
            print("User interrupted..")

        logger.info(
            f"{self.num_epochs_saved} {plural(self.num_epochs_saved, 'epoch')} "
            f"saved to '{self.results_file}'."
        )

        if self.current_best_epoch:
            HyperoptTools.try_export_params(
                self.config,
                self.hyperopter.get_strategy_name(),
                self.current_best_epoch,
            )

            HyperoptTools.show_epoch_details(
                self.current_best_epoch, self.total_epochs, self.print_json
            )
        elif self.num_epochs_saved > 0:
            print(
                f"No good result found for given optimization function in {self.num_epochs_saved} "
                f"{plural(self.num_epochs_saved, 'epoch')}."
            )
        else:
            # This is printed when Ctrl+C is pressed quickly, before first epochs have
            # a chance to be evaluated.
            print("No epochs evaluated yet, no best result.")