File size: 9,282 Bytes
13fc29d | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 | """
Run full pipeline: Stage 1 (sensors -> Farquhar -> A labels), validation, Stage 2 (IMS + labels -> train/eval), plots.
Requires: Data/Seymour/sensors_wide.csv, Data/ims/ims_merged_15min.csv (or run download_ims_data first).
Usage:
python -m scripts.run_pipeline
"""
from __future__ import annotations
import sys
from pathlib import Path
PROJECT_ROOT = Path(__file__).resolve().parent.parent
if str(PROJECT_ROOT) not in sys.path:
sys.path.insert(0, str(PROJECT_ROOT))
try:
from dotenv import load_dotenv
load_dotenv(PROJECT_ROOT / ".env")
except ImportError:
pass
import pandas as pd
def run_stage1():
"""Load sensors, filter daytime, compute A with Farquhar, save labels. Returns (df_sensor, labels)."""
from config import settings
from src.sensor_data_loader import SensorDataLoader
from src.farquhar_model import FarquharModel
settings.PROCESSED_DIR.mkdir(parents=True, exist_ok=True)
loader = SensorDataLoader()
df = loader.load()
df = loader.filter_daytime(df, par_threshold=50.0)
# Keep growing season only (May–September); vine dormant Oct–April, no photosynthesis
ts_col = "time" if "time" in df.columns else df.columns[0]
t = pd.to_datetime(df[ts_col], utc=True)
df = df.loc[t.dt.month.isin(settings.GROWING_SEASON_MONTHS)].copy()
if df.empty:
return None, None
model = FarquharModel()
A = model.compute_all(df)
# Index labels by timestamp; round to 15min and aggregate to align with IMS for Stage 2 merge
ts_col = "time" if "time" in df.columns else df.columns[0]
t = pd.to_datetime(df[ts_col], utc=True)
A = A.copy()
A.index = t.dt.floor("15min")
A = A.groupby(level=0).mean()
A.name = "A"
path = settings.PROCESSED_DIR / "stage1_labels.csv"
A.to_csv(path)
print(f"Stage 1: saved {len(A)} labels to {path}")
return df, A
def validate_stage1(df_sensor, labels, outputs_dir: Path):
"""Sanity check A (0–25 typical), diurnal plot, correlation with PAR and T."""
import numpy as np
import pandas as pd
if labels is None or labels.empty:
return
outputs_dir.mkdir(parents=True, exist_ok=True)
A = labels.dropna()
stats = {"count": len(A), "mean": A.mean(), "std": A.std(), "min": A.min(), "max": A.max()}
print("Stage 1 validation – A (umol m-2 s-1):", stats)
if A.max() > 50 or A.min() < -5:
print(" Note: typical grapevine A is 0–25; check inputs if out of range.")
try:
import matplotlib.pyplot as plt
fig, axes = plt.subplots(2, 1, figsize=(10, 8))
# Diurnal: A vs hour
ts = pd.to_datetime(A.index, utc=True)
hour = ts.hour + ts.minute / 60
axes[0].scatter(hour, A.values, alpha=0.3, s=5)
axes[0].set_xlabel("Hour (UTC)")
axes[0].set_ylabel("A (umol m-2 s-1)")
axes[0].set_title("Diurnal pattern (Stage 1)")
axes[0].set_xlim(0, 24)
if df_sensor is not None and not df_sensor.empty and "Air1_PAR_ref" in df_sensor.columns:
ts_col = "time" if "time" in df_sensor.columns else df_sensor.columns[0]
par_series = df_sensor.set_index(ts_col)["Air1_PAR_ref"]
common = A.reindex(par_series.index).dropna()
par = par_series.reindex(common.index).dropna()
common = common.reindex(par.index).dropna()
if len(par) > 10:
axes[1].scatter(par.values, common.values, alpha=0.3, s=5)
axes[1].set_xlabel("PAR (umol m-2 s-1)")
axes[1].set_ylabel("A (umol m-2 s-1)")
axes[1].set_title("A vs PAR")
plt.tight_layout()
plt.savefig(outputs_dir / "stage1_validation.png", dpi=150)
plt.close()
print(f" Saved {outputs_dir / 'stage1_validation.png'}")
except Exception as e:
print(" Plot skip:", e)
def run_stage2(outputs_dir: Path):
"""Load IMS + stage1 labels, preprocess, train, evaluate, save metrics and plots."""
from config import settings
from src.ims_client import IMSClient
from src.preprocessor import Preprocessor
from src.predictor import PhotosynthesisPredictor
import pandas as pd
import numpy as np
path_labels = settings.PROCESSED_DIR / "stage1_labels.csv"
if not path_labels.exists():
print("Stage 2: no stage1_labels.csv; run Stage 1 first.")
return
labels = pd.read_csv(path_labels, index_col=0, parse_dates=True).iloc[:, 0]
labels.index = pd.to_datetime(labels.index, utc=True)
labels = labels.dropna()
client = IMSClient()
ims_df = client.load_cached()
if ims_df.empty:
print("Stage 2: no IMS cache; run download_ims_data first.")
return
preproc = Preprocessor()
merged = preproc.merge_ims_with_labels(ims_df, labels, timestamp_index_labels=True)
if merged.empty:
print("Stage 2: no overlap between IMS and labels. Ensure IMS cache covers sensor date range (e.g. run download_ims_data --from YYYY-MM-DD --to YYYY-MM-DD for that range).")
return
merged = preproc.create_time_features(merged)
X_train, y_train, X_test, y_test = preproc.temporal_split(merged)
if X_train.empty or X_test.empty:
print("Stage 2: insufficient data after split.")
return
X_train_s = preproc.fit_transform_train(X_train)
X_test_s = preproc.transform_test(X_test)
# Ensure no sensor columns in features
assert "A" not in X_train_s.columns and "timestamp_utc" not in X_train_s.columns
predictor = PhotosynthesisPredictor()
predictor.train(X_train_s, y_train)
eval_df = predictor.evaluate(X_test_s, y_test)
print("Stage 2 metrics:")
print(eval_df.to_string(index=False))
eval_df.to_csv(outputs_dir / "stage2_metrics.csv", index=False)
predictor.plot_results(y_test, save_path=outputs_dir / "stage2_predictions.png")
imp = predictor.get_feature_importance()
if not imp.empty:
imp.to_csv(outputs_dir / "stage2_feature_importance.csv", index=False)
try:
import matplotlib.pyplot as plt
imp.set_index("feature")["importance"].plot(kind="barh", legend=False, figsize=(8, 4))
plt.tight_layout()
plt.savefig(outputs_dir / "stage2_feature_importance.png", dpi=150)
plt.close()
except Exception:
pass
print(f" Saved metrics and plots to {outputs_dir}")
def run_stage2_timeseries(outputs_dir: Path):
"""Load merged IMS+labels, build lag features, train/eval per horizon."""
from config import settings
from src.ims_client import IMSClient
from src.preprocessor import Preprocessor
from src.ts_predictor import TimeSeriesPredictor
import pandas as pd
path_labels = settings.PROCESSED_DIR / "stage1_labels.csv"
if not path_labels.exists():
print("TS: no stage1_labels.csv; run Stage 1 first.")
return
labels = pd.read_csv(path_labels, index_col=0, parse_dates=True).iloc[:, 0]
labels.index = pd.to_datetime(labels.index, utc=True)
labels = labels.dropna()
client = IMSClient()
ims_df = client.load_cached()
if ims_df.empty:
print("TS: no IMS cache; run download_ims_data first.")
return
preproc = Preprocessor()
merged = preproc.merge_ims_with_labels(ims_df, labels, timestamp_index_labels=True)
if merged.empty:
print("TS: no overlap between IMS and labels.")
return
merged = preproc.create_time_features(merged)
ts_pred = TimeSeriesPredictor()
merged = ts_pred.assign_season(merged)
merged = ts_pred.identify_sessions(merged)
merged = ts_pred.create_lag_features(merged)
print("Training time-series models per horizon (per season)...")
ts_metrics = ts_pred.train_all_horizons(merged)
# Save full per-season breakdown
ts_metrics.to_csv(outputs_dir / "ts_per_season_metrics.csv", index=False)
# Load cross-sectional baseline for comparison (overall only)
baseline_path = outputs_dir / "stage2_metrics.csv"
if baseline_path.exists():
baseline = pd.read_csv(baseline_path)
comparison = ts_pred.get_comparison_with_baseline(baseline)
else:
comparison = ts_metrics[ts_metrics["season"] == "all"].copy()
out_path = outputs_dir / "ts_horizon_comparison.csv"
comparison.to_csv(out_path, index=False)
print(f" Saved {out_path}")
print(f" Saved {outputs_dir / 'ts_per_season_metrics.csv'}")
def main():
from config import settings
outputs_dir = settings.OUTPUTS_DIR
outputs_dir.mkdir(parents=True, exist_ok=True)
print("Running Stage 1 (sensors -> Farquhar -> labels)...")
try:
df_sensor, labels = run_stage1()
except FileNotFoundError as e:
print(f"Stage 1 skipped (missing data): {e}")
df_sensor, labels = None, None
except Exception as e:
print(f"Stage 1 error: {e}")
raise
if labels is not None:
print("Stage 1 validation...")
validate_stage1(df_sensor, labels, outputs_dir)
print("Running Stage 2 (IMS + labels -> train/eval)...")
run_stage2(outputs_dir)
print("Running Stage 2 time-series forecasting...")
run_stage2_timeseries(outputs_dir)
print("Done.")
if __name__ == "__main__":
main()
|