def download() -> None:
"""
Download M4 dataset if doesn't exist.
"""
if os.path.isdir(DATASET_PATH):
logging.info(f'skip: {DATASET_PATH} directory already exists.')
return
download(INFO_URL, INFO_FILE_PATH)
m4_ids = pd.read_csv(INFO_FILE_PATH).M4id.values
def build_cache(files: str, cache_path: str) -> None:
timeseries_dict = OrderedDict(list(zip(m4_ids,
[[]] * len(m4_ids))))
logging.info(f'Caching {files}')
for train_csv in tqdm(glob(os.path.join(DATASET_PATH, files))):
dataset = pd.read_csv(train_csv)
dataset.set_index(dataset.columns[0], inplace=True)
for m4id, row in dataset.iterrows():
values = row.values
timeseries_dict[m4id] = values[~np.isnan(values)]
np.array(list(timeseries_dict.values())).dump(cache_path)
download(TRAINING_DATASET_URL, TRAINING_DATASET_FILE_PATH)
patoolib.extract_archive(TRAINING_DATASET_FILE_PATH,
outdir=DATASET_PATH)
build_cache('*-train.csv', TRAINING_DATASET_CACHE_FILE_PATH)
download(TEST_DATASET_URL, TEST_DATASET_FILE_PATH)
patoolib.extract_archive(TEST_DATASET_FILE_PATH, outdir=DATASET_PATH)
build_cache('*-test.csv', TEST_DATASET_CACHE_FILE_PATH)
naive2_archive = os.path.join(DATASET_PATH,
url_file_name(NAIVE2_FORECAST_URL))
download(NAIVE2_FORECAST_URL, naive2_archive)
patoolib.extract_archive(naive2_archive, outdir=DATASET_PATH)
"""
import logging
import os
from dataclasses import dataclass
import numpy as np
import pandas as pd
import patoolib
from common.http_utils import download, url_file_name
from common.settings import DATASETS_PATH
DATASET_URL = 'https://robjhyndman.com/data/27-3-Athanasopoulos1.zip'
DATASET_PATH = os.path.join(DATASETS_PATH, 'tourism')
DATASET_FILE_PATH = os.path.join(DATASET_PATH, url_file_name(DATASET_URL))
@dataclass()
class TourismMeta:
seasonal_patterns = ['Yearly', 'Quarterly', 'Monthly']
horizons = [4, 8, 24]
frequency = [1, 4, 12]
horizons_map = {'Yearly': 4, 'Quarterly': 8, 'Monthly': 24}
frequency_map = {'Yearly': 1, 'Quarterly': 4, 'Monthly': 12}
@dataclass()
class TourismDataset:
ids: np.ndarray
groups: np.ndarray
"""
M3 summary unit test
"""
import os
import unittest
import numpy as np
import pandas as pd
from common.http_utils import download, url_file_name
from common.settings import TESTS_STORAGE_PATH
from summary.m3 import M3Summary
FORECASTS_URL = 'https://forecasters.org/data/m3comp/M3Forecast.xls'
FORECASTS_FILE_PATH = os.path.join(TESTS_STORAGE_PATH, 'm3', url_file_name(FORECASTS_URL))
class TestM3Summary(unittest.TestCase):
def setUp(self) -> None:
download(FORECASTS_URL, FORECASTS_FILE_PATH)
def test_summary(self):
summary = M3Summary()
naive2 = pd.read_excel(FORECASTS_FILE_PATH, sheet_name='NAIVE2', header=None)
naive2_forecast = np.array([ts[~np.isnan(ts)]
for ts in naive2[naive2.columns[2:]].values])
result = summary.evaluate(naive2_forecast)
# based on http://www.forecastingprinciples.com/paperpdf/Makridakia-The%20M3%20Competition.pdf
# Tables 13-16 and Table 6 for Average.
self.assertEqual(result['M3Year'], 17.88)
def setUp(self) -> None:
winner_archive = os.path.join(TEST_STORAGE_PATH,
url_file_name(WINNER_FORECAST_URL))
download(WINNER_FORECAST_URL, winner_archive)
if not os.path.isfile(WINNER_FORECAST_PATH):
patoolib.extract_archive(winner_archive, outdir=TEST_STORAGE_PATH)
import pandas as pd
import patoolib
from tqdm import tqdm
from common.http_utils import download, url_file_name
from common.settings import DATASETS_PATH
TRAINING_DATASET_URL = 'https://www.m4.unic.ac.cy/wp-content/uploads/2017/12/M4DataSet.zip'
TEST_DATASET_URL = 'https://www.m4.unic.ac.cy/wp-content/uploads/2018/07/M-test-set.zip'
INFO_URL = 'https://www.m4.unic.ac.cy/wp-content/uploads/2018/12/M4Info.csv'
NAIVE2_FORECAST_URL = 'https://github.com/M4Competition/M4-methods/raw/master/Point%20Forecasts/submission-Naive2.rar'
DATASET_PATH = os.path.join(DATASETS_PATH, 'm4')
TRAINING_DATASET_FILE_PATH = os.path.join(DATASET_PATH,
url_file_name(TRAINING_DATASET_URL))
TEST_DATASET_FILE_PATH = os.path.join(DATASET_PATH,
url_file_name(TEST_DATASET_URL))
INFO_FILE_PATH = os.path.join(DATASET_PATH, url_file_name(INFO_URL))
NAIVE2_FORECAST_FILE_PATH = os.path.join(DATASET_PATH, 'submission-Naive2.csv')
TRAINING_DATASET_CACHE_FILE_PATH = os.path.join(DATASET_PATH, 'training.npz')
TEST_DATASET_CACHE_FILE_PATH = os.path.join(DATASET_PATH, 'test.npz')
@dataclass()
class M4Dataset:
ids: np.ndarray
groups: np.ndarray
frequencies: np.ndarray
horizons: np.ndarray