model_accuracy/server/LEMON-master/scripts/mutation/model_mutation_generators.py

import sys

sys.path.append("../")
from scripts.mutation.model_mutation_operators import *
import scripts.mutation.utils as utils
import scripts.tools.utils as utils_tools
import argparse

warnings.filterwarnings("ignore")
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "2"  # 只显示 warning 和 Error
# os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
# os.environ["CUDA_VISIBLE_DEVICES"] = ""
mylogger = Logger()
import tensorflow as tf
import psutil


def generate_model_by_model_mutation(model, operator, mutate_ratio=0.3):
    """
    Generate models using specific mutate operator
    :param model: model loaded by keras (tensorflow backend default)
    :param operator: mutation operator
    :param mutate_ratio: ratio of selected neurons
    :return: mutation model object
    """
    if operator == "WS":
        mutate_indices = utils.ModelUtils.weighted_layer_indices(model)
        mylogger.info("Generating model using {}".format(operator))
        return WS_mut(
            model=model,
            mutation_ratio=mutate_ratio,
            mutated_layer_indices=mutate_indices,
        )
    elif operator == "GF":
        mylogger.info("Generating model using {}".format(operator))
        return GF_mut(model=model, mutation_ratio=mutate_ratio)
    elif operator == "NEB":
        mylogger.info("Generating model using {}".format(operator))
        return NEB_mut(model=model, mutation_ratio=mutate_ratio)
    elif operator == "NAI":
        mylogger.info("Generating model using {}".format(operator))
        return NAI_mut(model=model, mutation_ratio=mutate_ratio)
    elif operator == "NS":
        mylogger.info("Generating model using {}".format(operator))
        return NS_mut(model=model)
    elif operator == "ARem":
        mylogger.info("Generating model using {}".format(operator))
        return ARem_mut(model=model)
    elif operator == "ARep":
        mylogger.info("Generating model using {}".format(operator))
        return ARep_mut(model=model)
    elif operator == "LA":
        mylogger.info("Generating model using {}".format(operator))
        return LA_mut(model=model)
    elif operator == "LC":
        mylogger.info("Generating model using {}".format(operator))
        return LC_mut(model=model)
    elif operator == "LR":
        mylogger.info("Generating model using {}".format(operator))
        return LR_mut(model=model)
    elif operator == "LS":
        mylogger.info("Generating model using {}".format(operator))
        return LS_mut(model=model)
    elif operator == "MLA":
        mylogger.info("Generating model using {}".format(operator))
        return MLA_mut(model=model)
    else:
        mylogger.info("No such Mutation operator {}".format(operator))
        return None


def all_mutate_ops():
    return [
        "WS",
        "GF",
        "NEB",
        "NAI",
        "NS",
        "ARem",
        "ARep",
        "LA",
        "LC",
        "LR",
        "LS",
        "MLA",
    ]


if __name__ == "__main__":

    """Parser of command args"""
    parse = argparse.ArgumentParser()  # 创建解析器
    parse.add_argument("--model", type=str, help="model path")
    parse.add_argument("--mutate_op", type=str, help="model mutation operator")
    parse.add_argument("--save_path", type=str, help="model save path")
    parse.add_argument("--mutate_ratio", type=float, help="mutate ratio")
    # 添加参数
    flags, unparsed = parse.parse_known_args(
        sys.argv[1:]
    )  # 解析参数，flags接收conf文件里的参数

    # model = r"D:\LEMON\LEMON-master\origin_model\lenet5-mnist_origin.h5"
    # mutate_op = r'NS' #变异算子
    # save_path = r"/lemon_outputs"
    # mutate_ratio = 0.3

    import keras

    model_path = flags.model
    mutate_ratio = flags.mutate_ratio
    print("Current {}; Mutate ratio {}".format(flags.mutate_op, mutate_ratio))
    origin_model = keras.models.load_model(
        model_path, custom_objects=utils.ModelUtils.custom_objects()
    )
    mutated_model = generate_model_by_model_mutation(
        model=origin_model, operator=flags.mutate_op, mutate_ratio=mutate_ratio
    )

    if mutated_model is None:
        raise Exception("Error: Model mutation using {} failed".format(flags.mutate_op))
    else:

        mutated_model.save(flags.save_path)