ampli/py/predict.py

from argparse import ArgumentParser, FileType
import numpy as np
import pandas as p
import statsmodels.formula.api as smf
import datetime as dt
import pickle
from sys import stdout
from pymodels import thirtyoffset, predweather, harmonic
from pprint import pprint
from util import datevalid
from signal import signal, SIGPIPE, SIG_DFL

def preddemand(model, weather, harmonics):
    harmstart = thirtyoffset(weather.index.min())
    utv = weather.index
    harmlen = len(utv)

    yharm = harmonic(harmlen, period=365.25 * 48, start=harmstart,
                     harmonics=harmonics[0], prefix="y")
    yharm.index = utv

    wharm = harmonic(harmlen, period=7 * 48, start=harmstart,
                     harmonics=harmonics[1], prefix="w")
    wharm.index = utv

    dharm = harmonic(harmlen, period=48, start=harmstart,
                     harmonics=harmonics[2], prefix="d")
    dharm.index = utv

    df = p.concat([yharm, wharm, dharm, weather], axis = 1)
    return model.predict(df)

def main():
    parser = ArgumentParser(description=
                            'Predict from harmonic model of cluster')
    parser.add_argument("-m", "--model-file", dest="model_file",
                        help="filename for models",
                        required = True,
                        type=FileType('rb'))
    parser.add_argument("-w", "--weather",  dest="weather_file",
                        help="input weather pickle path",
                        required=False,
                        type=FileType('rb'))
    parser.add_argument("-o", "--output", dest="output_file",
                        help="file to save result (default stdout)")
    parser.add_argument("-t", "--temperature", dest="temp",
                        help = "min and max temperature, if not using "
                        "weather dataset, e.g. 2.0 10.5", 
                        required=False,
                        type=float, nargs=2)
    parser.add_argument("-s", "--start-date", 
                        dest = "startdate", 
                        help = "start date for prediction; format: YYYY-MM-DD", 
                        metavar="START_DATE", 
                        required = True,
                        type = datevalid)
    parser.add_argument("-e", "--end-date", 
                        dest = "enddate", 
                        help = "end date for prediction; format: YYYY-MM-DD", 
                        metavar="END_DATE", 
                        required = True,
                        type = datevalid)
    parser.add_argument("-c", "--cluster",
                        dest = "cluster",
                        help = "cluster to predict for",
                        type = int,
                        required = True)
    parser.add_argument("--pkl",
                        help="output as pkl rather than csv",
                        dest = "csv",
                        action="store_false")
    args = parser.parse_args()

    if args.temp is None and args.weather_file is None:
        parser.error("Either the temperature range or a weather "
                     "dataset must be specified")

    mods = pickle.load(args.model_file)

    if args.cluster not in mods["clusters"]:
        parser.error(f"cluster ('{args.cluster}') not in model")

    if args.output_file is None or args.output_file == "-":
        args.output_file = stdout

    wdat = []
    
    if args.weather_file is not None:
        weather = p.read_pickle(args.weather_file)
        if (args.startdate < weather['temp_timestamp'].min() or 
            args.enddate > weather['temp_timestamp'].max()):
            parser.error("Start and or end date not in supplied weather dataset")

        wantedtimes = p.date_range(args.startdate,
                                   args.enddate, freq="30 min")
        wantedtimes = p.Series(wantedtimes.round(freq="30 min"))
        wanteddf = p.DataFrame(index=wantedtimes)
        weather = wanteddf.join(weather.set_index(['temp_timestamp']),
                                how='left').drop('temp_date', axis=1)
        weather.index.name = "read_time"
        weather = weather.fillna(method='ffill', axis=0)

        # Add rolling columns
        weather['min_rolling'] = weather['tmin_c'].rolling("1D").min()
        weather['max_rolling'] = weather['tmax_c'].rolling("1D").max()

        wharmstart = thirtyoffset(weather.index.min())
        wharm = harmonic(weather.shape[0], period=365.25 * 48, start=wharmstart,
                         harmonics=mods["weather_harmonics"])
        wharm.index = wantedtimes
        pred_max = weather['max_rolling'] - mods["max_temp"].predict(wharm)
        pred_min = weather['min_rolling'] - mods["min_temp"].predict(wharm)
        wdat = p.DataFrame({
            "max_resid": pred_max,
            "min_resid": pred_min
        })

    else:
        mint = min(args.temp)
        maxt = max(args.temp)
        pred_max = maxt - predweather(mods["max_temp"],
                               args.startdate,
                               args.enddate,
                               harmonics = mods["weather_harmonics"])
        pred_min = mint - predweather(mods["min_temp"],
                               args.startdate,
                               args.enddate,
                               harmonics = mods["weather_harmonics"])
        wdat = p.DataFrame({
            "max_resid": pred_max,
            "min_resid": pred_min
        })


    dpred = p.DataFrame(
        {
            "predicted": preddemand(mods["icp"][args.cluster], wdat, mods["icp_harmonics"])
        })

    if args.csv:
        signal(SIGPIPE, SIG_DFL) # Prevent broken pipe errors when piping to less or head
        dpred.to_csv(args.output_file)
    else:
        dpred.to_pickle(args.output_file)


    args.model_file.close()
    if args.weather_file is not None:
        args.weather_file.close()


if __name__ == "__main__":
    main()