Spaces:
Sleeping
Sleeping
File size: 8,208 Bytes
b099c5d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 |
from typing import Dict
import warnings
from pathlib import Path
import pandas as pd
from typing import Any
from tqdm import tqdm
import numpy as np
from utils.metrics import mae, bias
from statsforecast import StatsForecast
from statsforecast.models import (
# Baselines
Naive,
SeasonalNaive,
RandomWalkWithDrift,
HistoricAverage,
WindowAverage,
# Exponential smoothing / Holt / Holt-Winters
SimpleExponentialSmoothingOptimized,
SeasonalExponentialSmoothingOptimized,
Holt,
HoltWinters,
# Theta family
Theta,
OptimizedTheta,
DynamicTheta,
DynamicOptimizedTheta,
# ARIMA
# AutoARIMA,
# Intermittent
CrostonClassic,
CrostonOptimized,
CrostonSBA,
)
### ----- Configuration ----- ###
TRAIN_DATA_PATH = Path("data/processed/train.csv")
TEST_DATA_PATH = Path("data/processed/test.csv")
METRICS_PATH = Path("metrics/baseline_metrics.csv")
PREDICTIONS_PATH = Path("metrics/baseline_predictions.csv")
### ------------------------- ###
HORIZON = 14 # Forecast horizon
warnings.filterwarnings("ignore")
# Adding type hints for better code clarity and numpy style comments for documentation
def build_baseline_models(
season_length: int = 7,
window_size: int = 4,
) -> Dict[str, Any]:
"""Build a dictionary of baseline forecasting models.
Parameters
----------
season_length : int, optional
Seasonality length, by default 7
window_size : int, optional
Window size for moving average models, by default 4
Returns
-------
Dict[str, Any]
Dictionary of baseline forecasting models
"""
models= {
# ----------------------
# 1) Naive family
# ----------------------
str(Naive().__class__.__name__): Naive(),
str(SeasonalNaive(season_length=season_length).__class__.__name__): SeasonalNaive(season_length=season_length),
str(RandomWalkWithDrift().__class__.__name__): RandomWalkWithDrift(),
str(HistoricAverage().__class__.__name__): HistoricAverage(),
str(WindowAverage(window_size=window_size).__class__.__name__): WindowAverage(window_size=window_size),
# ----------------------
# 2) SES / Holt / Holt-Winters
# ----------------------
# SES ~ simple exponential smoothing
str(SimpleExponentialSmoothingOptimized().__class__.__name__): SimpleExponentialSmoothingOptimized(),
str(SeasonalExponentialSmoothingOptimized(season_length=season_length).__class__.__name__): SeasonalExponentialSmoothingOptimized(season_length=season_length),
# Holt: level + trend, no seasonality
str(Holt().__class__.__name__): Holt(),
str(HoltWinters(
season_length=7, # e.g. weekly seasonality for daily data # or "multiplicative"
).__class__.__name__): HoltWinters(
season_length=7, # e.g. weekly seasonality for daily data
),
# ----------------------
# 3) Theta family
# ----------------------
str(Theta().__class__.__name__): Theta(),
str(OptimizedTheta().__class__.__name__): OptimizedTheta(),
str(DynamicTheta().__class__.__name__): DynamicTheta(),
str(DynamicOptimizedTheta().__class__.__name__): DynamicOptimizedTheta(),
# ----------------------
# 4) ARIMA baseline
# ----------------------
# str(AutoARIMA().__class__.__name__): AutoARIMA(season_length=season_length),
# ----------------------
# 5) Intermittent demand
# ----------------------
str(CrostonClassic().__class__.__name__): CrostonClassic(),
str(CrostonSBA().__class__.__name__): CrostonSBA(),
str(CrostonOptimized().__class__.__name__): CrostonOptimized(),
}
return models
# Adding type hints for better code clarity and numpy style comments for documentation
def compute_baseline_forecasts(
df: pd.DataFrame,
models: Dict[str, Any],
horizon: int = HORIZON,
) -> pd.DataFrame:
"""
df: train dataframe with columns ['id', 'date', 'sales']
returns: long dataframe with columns
['id', 'model', 'h', 'forecast']
"""
results = []
sku_ids = df['id'].unique()
for sku_id in tqdm(sku_ids):
sku_data = df[df['id'] == sku_id].sort_values('date').copy()
if len(sku_data) <= horizon + 5:
continue
sku_data.rename(columns={'sales': 'y', 'id': 'unique_id'}, inplace=True)
# dummy calendar for StatsForecast — only order matters
sku_data['ds'] = pd.date_range(start='2021-01-01', periods=len(sku_data), freq='D')
sku_data = sku_data[['unique_id', 'ds', 'y']]
for model_name, model in models.items():
sf = StatsForecast(models=[model], freq='D', n_jobs=1)
sf.fit(sku_data)
forecast_df = sf.predict(h=horizon)
# StatsForecast sometimes uses short aliases for columns
# map model_name → column name
col_map = {
"RandomWalkWithDrift": "RWD",
"SimpleExponentialSmoothingOptimized": "SESOpt",
"SeasonalExponentialSmoothingOptimized": "SeasESOpt",
}
col = col_map.get(model_name, model.__class__.__name__)
if col not in forecast_df.columns:
raise ValueError(f"Column {col} not found for model {model_name}")
forecast_values = forecast_df[col].values
for step in range(horizon):
results.append({
"id": sku_id,
"model": model_name,
"h": step + 1,
"forecast": float(forecast_values[step]),
})
return pd.DataFrame(results)
def compute_metrics(
test_df: pd.DataFrame,
forecasts_df: pd.DataFrame,
horizon: int = HORIZON,
) -> pd.DataFrame:
"""
test_df: ['id', 'date', 'sales', ...]
forecasts_df: ['id', 'model', 'h', 'forecast']
returns: ['id', 'model', 'mae', 'bias', 'score']
"""
test_df = test_df.sort_values(["id", "date"]).copy()
# assign step index 1..H per SKU in time order
test_df["h"] = test_df.groupby("id").cumcount() + 1
metrics_rows = []
for (sku_id, model_name), g_fore in forecasts_df.groupby(["id", "model"]):
g_test = test_df[test_df["id"] == sku_id].copy()
if g_test["h"].max() < horizon:
# test shorter than horizon for some reason
continue
# align on h
merged = pd.merge(
g_test[["id", "h", "sales"]],
g_fore[["id", "h", "forecast"]],
on=["id", "h"],
how="inner",
)
if merged.empty:
continue
y_true = merged["sales"].values
y_pred = merged["forecast"].values
m = mae(y_true, y_pred)
b = bias(y_true, y_pred)
s = m + abs(b)
metrics_rows.append({
"id": sku_id,
"model": model_name,
"mae": float(m),
"bias": float(b),
"score": float(s),
})
return pd.DataFrame(metrics_rows)
if __name__ == "__main__":
# Load data
train_df = pd.read_csv(TRAIN_DATA_PATH)
test_df = pd.read_csv(TEST_DATA_PATH)
# Build baseline models
baseline_models = build_baseline_models()
# Forecast from train into test horizon
train_forecasts = compute_baseline_forecasts(train_df, baseline_models, horizon=HORIZON)
# Save raw forecasts (for later UI)
PREDICTIONS_PATH.parent.mkdir(parents=True, exist_ok=True)
train_forecasts.to_csv(PREDICTIONS_PATH, index=False)
# Compute metrics vs test
metrics_df = compute_metrics(test_df, train_forecasts, horizon=HORIZON)
metrics_df.to_csv(METRICS_PATH, index=False)
print("Saved:")
print(f" - forecasts → {PREDICTIONS_PATH}")
print(f" - metrics → {METRICS_PATH}")
|