コード例 #1
0
def get_best_global_scenarios(quantile_step):
    db = DBAccess(env.DB_RESULT_NAME)
    db.clear_collection(env.DB_DETAILED_BEST_RESULT_COLLECTION_NAME)

    scenarios = db.get_records(env.DB_GLOBAL_RESULT_COLLECTION_NAME,
                               {}).sort([("Cost PV", DESCENDING)])
    step = int(quantile_step * scenarios.count())
    representative_scenarios = [
        scenarios.skip(step * i)[0]
        for i in range(0, int(scenarios.count() / step))
    ]
    db.save_to_db_no_check(env.DB_GLOBAL_BEST_RESULT_COLLECTION_NAME,
                           representative_scenarios)
コード例 #2
0
def get_best_detailed_scenarios(quantile_step):
    db = DBAccess(env.DB_RESULT_NAME)
    db.clear_collection(env.DB_DETAILED_BEST_RESULT_COLLECTION_NAME)

    scenarios = db.get_fields(env.DB_GLOBAL_RESULT_COLLECTION_NAME, {
        "Cost PV": 1,
        "Scenario": 1
    }, [("Cost PV", DESCENDING)])
    step = int(quantile_step * scenarios.count())
    points = [
        scenarios.skip(step * i)[0]["Scenario"]
        for i in range(0, int(scenarios.count() / step))
    ]
    representative_scenarios = db.get_records(
        env.DB_DETAILED_RESULT_COLLECTION_NAME, {"Scenario": {
            "$in": points
        }})
    db.save_to_db_no_check(env.DB_DETAILED_BEST_RESULT_COLLECTION_NAME,
                           representative_scenarios)
コード例 #3
0
def get_best_scenarios(quantile_step, db_name="mine2farm"):
    update_cache(db_name, -1)
    try:
        time_start = datetime.datetime.now().strftime("%d/%m/%y %H:%M:%S")

        # insert status of best scenarios "running"
        db_history = DBAccess(env.MONITORING_DB_NAME)
        query_insert = {
            'time_start': time_start,
            'db_name': db_name,
            'quantile_step': quantile_step,
            'status': -1
        }
        _id = db_history.save_to_db_no_check(
            env.MONITORING_COLLECTION_HISTORY_BEST_NAME, query_insert)

        # get best representative scenarios
        quantile_step = quantile_step / 100.
        reset_db_name(db_name)
        db = DBAccess(env.DB_RESULT_NAME)
        logger.info("Deleting best collections from DB")
        db.clear_collection(env.DB_GLOBAL_BEST_RESULT_COLLECTION_NAME)
        db.clear_collection(env.DB_DETAILED_BEST_RESULT_COLLECTION_NAME)
        scenarios = db.get_records(env.DB_GLOBAL_RESULT_COLLECTION_NAME,
                                   {}).sort([("Cost PV", DESCENDING)])

        scenarios_count = scenarios.count()
        step = int(quantile_step * scenarios_count)
        # save to db
        if step == 0:
            # all scenarios are concerned
            logger.info("Moving all scenarios to best collections")
            db.copy_to_collection(env.DB_GLOBAL_RESULT_COLLECTION_NAME,
                                  env.DB_GLOBAL_BEST_RESULT_COLLECTION_NAME)
            db.copy_to_collection(env.DB_DETAILED_RESULT_COLLECTION_NAME,
                                  env.DB_DETAILED_BEST_RESULT_COLLECTION_NAME)
            details_count = db.count(
                env.DB_DETAILED_BEST_RESULT_COLLECTION_NAME)
        else:
            # filter on specific scenarios
            representative_scenario_ids = [
                scenarios.skip(step * i)[0]["Scenario"]
                for i in range(0, int(scenarios_count / step))
            ]
            logger.info("List of selected best scenarios: %s" %
                        representative_scenario_ids)
            # simulate
            scenarios_global, scenarios_details = \
                Simulator().simulate(scenarios_filter=representative_scenario_ids, logistics_lp=env.LOGISTICS_LP)
            # save
            for scenario in scenarios_global:
                db.save_to_db_no_check(
                    env.DB_GLOBAL_BEST_RESULT_COLLECTION_NAME,
                    scenarios_global[scenario])
            for scenario in scenarios_details:
                json_data = json.dumps(NodeJSONEncoder().encode(
                    scenarios_details[scenario]))
                data = json.loads(json.loads(json_data))
                db.save_to_db_no_check(
                    env.DB_DETAILED_BEST_RESULT_COLLECTION_NAME, data)
            details_count = len(scenarios_details)

        # status update
        query_insert['global_count'] = scenarios_count
        query_insert['detailed_count'] = details_count
        filter_ = {'_id': ObjectId(_id)}
        db_history.update_record(
            collection=env.MONITORING_COLLECTION_HISTORY_BEST_NAME,
            filter_=filter_,
            data=query_insert)

        # raw materials sensitivities
        logger.info("Running sensitivity over raw materials")
        db.clear_collection(env.DB_SENSITIVITY_COLLECTION_NAME)
        raw_materials_df = Driver().get_data("raw_materials")
        shocks = {}
        for raw_material in raw_materials_df:
            item = raw_material["Item"]
            shocks[item] = 1
        scenarios_df = pd.DataFrame(Driver().get_results(
            env.DB_GLOBAL_BEST_RESULT_COLLECTION_NAME))
        scenarios_dic = Utils.get_scenario_from_df(scenarios_df)
        risk_engine = RiskEngine()

        for scenario_id in scenarios_dic:
            deltas = risk_engine.compute_delta(scenarios_dic[scenario_id],
                                               shocks,
                                               with_logistics=env.LOGISTICS_LP)

            deltas['Scenario'] = int(scenario_id)
            db.save_to_db_no_check(env.DB_SENSITIVITY_COLLECTION_NAME, deltas)

        # status update
        query_insert['time_end'] = datetime.datetime.now().strftime(
            "%d/%m/%y %H:%M:%S")
        query_insert['status'] = 0
        filter_ = {'_id': ObjectId(_id)}
        db_history.update_record(
            collection=env.MONITORING_COLLECTION_HISTORY_BEST_NAME,
            filter_=filter_,
            data=query_insert)
        update_cache(db_name, 0)

    except Exception as e:
        logger.error("Best scenarios failed")
        update_cache(db_name, 0)