Broken DAG: [/usr/local/airflow/dags/my_dag.py] No module named 'airflow.operators.subdag'

Question

I'm running airflow inside docker container and getting airflow image (puckel/docker-airflow:latest) from docker hub. I can access Airflow UI through localhost:8080 but without executing the DAG and the error mentioned in the subject above. I'm even writing pip command to install apache-airflow in my Dockerfile. Here is how my Dockerfile, docker-compose.yml and dag.py looks like:

Dockerfile:

FROM puckel/docker-airflow:latest
RUN pip install requests
RUN pip install pandas
RUN pip install 'apache-airflow'

docker-compose.yml:

version: '3.7'
services:

    redis:
        image: redis:5.0.5
        environment:
            REDIS_HOST: redis
            REDIS_PORT: 6379
        ports:
          - 6379:6379

    postgres:
        image: postgres:9.6
        environment:
          - POSTGRES_USER=airflow
          - POSTGRES_PASSWORD=airflow
          - POSTGRES_DB=airflow
          - PGDATA=/var/lib/postgresql/data/pgdata
        volumes:  
          - ./pgdata:/var/lib/postgresql/data/pgdata
        logging:
          options:
            max-size: 10m
            max-file: "3"

      webserver:
        build: ./dockerfiles
        restart: always
        depends_on:
          - postgres
          - redis
        environment:
          - LOAD_EX=n
          - FERNET_KEY=46BKJoQYlPPOexq0OhDZnIlNepKFf87WFwLbfzqDDho=
          - EXECUTOR=Celery
        logging:
          options:
            max-size: 10m
            max-file: "3"
        volumes:
          - ./dags:/usr/local/airflow/dags
          - ./config/airflow.cfg:/usr/local/airflow/airflow.cfg
        ports:
          - "8080:8080"
        command: webserver
        healthcheck:
          test: ["CMD-SHELL", "[ -f /usr/local/airflow/airflow-webserver.pid ]"]
          interval: 30s
          timeout: 30s
          retries: 3

      flower:
        build: ./dockerfiles
        restart: always
        depends_on:
          - redis
        environment:
          - EXECUTOR=Celery
        ports:
          - "5555:5555"
        command: flower

      scheduler:
        build: ./dockerfiles
        restart: always
        depends_on:
          - webserver
        volumes:
          - ./dags:/usr/local/airflow/dags
        environment:
          - LOAD_EX=n
          - FERNET_KEY=46BKJoQYlPPOexq0OhDZnIlNepKFf87WFwLbfzqDDho=
          - EXECUTOR=Celery
        command: scheduler

      worker:
        build: ./dockerfiles
        restart: always
        depends_on:
          - scheduler
        volumes:
          - ./dags:/usr/local/airflow/dags
        environment:
          - FERNET_KEY=46BKJoQYlPPOexq0OhDZnIlNepKFf87WFwLbfzqDDho=
          - EXECUTOR=Celery
        command: worker

dag.py:

    from airflow import DAG
    from airflow.operators.subdag import SubDagOperator
    from airflow.operators.python import PythonOperator, BranchPythonOperator
    from airflow.operators.bash import BashOperator
    from datetime import datetime
    from random import randint

    def _choosing_best_model(ti):
        accuracies = ti.xcom_pull(task_ids=[
            'training_model_A',
            'training_model_B',
            'training_model_C'
            ])
        if max(accuracies) > 8:
            return 'accurate'
        return 'inaccurate'

    def _training_model(model):
        return randint(1, 10)

    with DAG("test",
        start_date=datetime(2021, 1 ,1), 
        schedule_interval='@daily', 
        catchup=False) as dag:

        training_model_tasks = [
            PythonOperator(
                task_id=f"training_model_{model_id}",
                python_callable=_training_model,
                op_kwargs={
                    "model": model_id
                }
            ) for model_id in ['A', 'B', 'C']
        ]

        choosing_best_model = BranchPythonOperator(
            task_id="choosing_best_model",
            python_callable=_choosing_best_model
        )

        accurate = BashOperator(
            task_id="accurate",
            bash_command="echo 'accurate'"
        )

        inaccurate = BashOperator(
            task_id="inaccurate",
            bash_command=" echo 'inaccurate'"
        )

        training_model_tasks >> choosing_best_model >> [accurate, inaccurate]

Am I missing anything here? Please let me know if you can. Thanks :)