sleonca's picture
Upload folder using huggingface_hub
b47743b verified
Raw History Blame Contribute Delete
1.2 kB
"""The submitted model: one forecaster per metric, chosen on pre-2019 backtests (02_backtest.py)."""
import pandas as pd
import models
SPECS = {
# mean of 15-yr level + 1.25 x pooled 50-yr warming rate, and the regional-anomaly model
"annual_avg_temp": [(models.level_trend, dict(k=15, L=50, slope_scale=1.25)),
(models.regional_anom, dict(k_nn=10, L=30, k_off=10, w_nat=0.5))],
# shifted daily weather from the last 10 and last 20 years, warmed to the target year
"cdd": [(models.cdd_shift, dict(N=10, L=50, slope_scale=1.25)),
(models.cdd_shift, dict(N=20, L=50, slope_scale=1.25))],
# 30-yr level + half of the regional (20 nearest cities) 70-yr trend
"hottest_day_temp": [(models.level_trend, dict(k=30, L=70, k_nn=20, slope_scale=0.5))],
# 50-yr level + half of the regional 50-yr trend
"total_precip_in": [(models.level_trend, dict(k=50, L=50, k_nn=20, slope_scale=0.5))],
}
def forecast(lv, origin, years, coords, metric):
ps = [fn(lv, origin, years, coords, metric, **kw).set_index(["ID", "metric", "Year"]).pred
for fn, kw in SPECS[metric]]
return (sum(ps) / len(ps)).rename("pred").reset_index()