Source code for gt4sd.training_pipelines.moses.organ.core

#
# MIT License
#
# Copyright (c) 2022 GT4SD team
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
#
"""Moses Organ training pipeline."""
import argparse
import ast
import logging
import os
from dataclasses import dataclass, field
from typing import Any, Dict

from guacamol_baselines.moses_baselines.organ_train import main
from moses.script_utils import MetricsReward

from ...core import TrainingPipelineArguments
from ..core import MosesTrainingArguments, MosesTrainingPipeline

logger = logging.getLogger(__name__)
logger.addHandler(logging.NullHandler())


[docs]class MosesOrganTrainingPipeline(MosesTrainingPipeline): """Moses Organ training pipelines."""
[docs] def train( # type: ignore self, training_args: Dict[str, Any], model_args: Dict[str, Any], dataset_args: Dict[str, Any], ) -> None: """Generic training function for Moses Organ training. Args: training_args: training arguments passed to the configuration. model_args: model arguments passed to the configuration. dataset_args: dataset arguments passed to the configuration. """ params = {**training_args, **model_args, **dataset_args} os.makedirs(os.path.dirname(params["model_save"]), exist_ok=True) os.makedirs(os.path.dirname(params["log_file"]), exist_ok=True) os.makedirs(os.path.dirname(params["config_save"]), exist_ok=True) os.makedirs(os.path.dirname(params["vocab_save"]), exist_ok=True) params["addition_rewards"] = list( map(str.strip, params["addition_rewards"].split(",")) ) params["discriminator_layers"] = ast.literal_eval( params["discriminator_layers"] ) args = argparse.Namespace(**params) main(args)
[docs]@dataclass class MosesOrganTrainingArguments(MosesTrainingArguments): """Arguments related to Moses Organ training.""" generator_pretrain_epochs: int = field( default=50, metadata={"help": "Number of epochs for generator pretraining."} ) discriminator_pretrain_epochs: int = field( default=50, metadata={"help": "Number of epochs for discriminator pretraining."} ) pg_iters: int = field( default=1000, metadata={"help": "Number of iterations for policy gradient training."}, ) n_batch: int = field(default=64, metadata={"help": "Size of batch."}) lr: float = field(default=1e-4, metadata={"help": "Learning rate."}) n_jobs: int = field(default=8, metadata={"help": "Number of threads."}) n_workers: int = field(default=8, metadata={"help": "Number of workers."}) clip_grad: int = field( default=5, metadata={"help": "Clip PG generator gradients to this value."} ) rollouts: int = field(default=16, metadata={"help": "Number of rollouts."}) generator_updates: int = field( default=1, metadata={"help": "Number of updates of generator per iteration."} ) discriminator_updates: int = field( default=1, metadata={"help": "Number of updates of discriminator per iteration."}, ) discriminator_epochs: int = field( default=10, metadata={"help": "Number of epochs of discriminator per iteration."}, ) reward_weight: float = field( default=0.7, metadata={"help": "Reward weight for policy gradient training."} ) addition_rewards: str = field( default="sa", metadata={ "help": f"Comma separated list of rewards. Feasible values from: {','.join(MetricsReward.supported_metrics)}. Defaults to optimization of SA." }, ) max_length: int = field( default=100, metadata={"help": "Maximum length for sequence."} ) n_ref_subsample: int = field( default=500, metadata={ "help": "Number of reference molecules (sampling from training data)." }, )
[docs]@dataclass class MosesOrganModelArguments(TrainingPipelineArguments): """Arguments related to Moses Organ model.""" __name__ = "model_args" embedding_size: int = field( default=32, metadata={"help": "Embedding size in generator and discriminator."} ) hidden_size: int = field( default=512, metadata={"help": "Size of hidden state for lstm layers in generator."}, ) num_layers: int = field( default=2, metadata={"help": "Number of lstm layers in generator."} ) dropout: float = field( default=0.0, metadata={"help": "Dropout probability for lstm layers in generator."}, ) discriminator_layers: str = field( default="[(100, 1), (200, 2), (200, 3), (200, 4), (200, 5), (100, 6), (100, 7), (100, 8), (100, 9), (100, 10), (160, 15), (160, 20)]", metadata={ "help": "String representation of numbers of features for convolutional layers in discriminator." }, ) discriminator_dropout: float = field( default=0.0, metadata={"help": "Dropout probability for discriminator."} )