BijliCast / app.py
muhammadumar15's picture
Add @spaces.GPU decorator for ZeroGPU startup
1f43f80 verified
Raw History Blame Contribute Delete
6.37 kB
"""BijliCast — Gradio front end for Hugging Face Spaces (ZeroGPU)."""
from __future__ import annotations
import json
import sys
from pathlib import Path
import matplotlib
matplotlib.use("Agg")
import gradio as gr # noqa: E402
import matplotlib.pyplot as plt # noqa: E402
import pandas as pd # noqa: E402
ROOT = Path(__file__).resolve().parent
if str(ROOT) not in sys.path:
sys.path.insert(0, str(ROOT))
from service import BijliCastService # noqa: E402
try: # available inside a ZeroGPU Space; optional locally
import spaces
gpu = spaces.GPU(duration=10)
except Exception: # noqa: BLE001 - local/dev fallback
def gpu(fn):
return fn
SERVICE = BijliCastService(ROOT)
HOUSES = SERVICE.houses()
HOUSE_IDS = [h["house_id"] for h in HOUSES]
BY_ID = {h["house_id"]: h for h in HOUSES}
INFO = SERVICE.model_info()
WEATHER_FIELDS = [
("temperature", "Temperature (°C)", -10.0, 55.0, 0.5),
("humidity", "Humidity (%)", 0.0, 100.0, 1.0),
("dew", "Dew point (°C)", -20.0, 40.0, 0.5),
("precipitation", "Precipitation (mm)", 0.0, 50.0, 0.1),
("wind_speed", "Wind speed (km/h)", 0.0, 120.0, 0.5),
("wind_direction", "Wind direction (°)", 0.0, 360.0, 1.0),
("pressure", "Pressure (hPa)", 900.0, 1050.0, 0.5),
("solar_radiation", "Solar radiation (W/m²)", 0.0, 1200.0, 5.0),
]
PROFILE_ROWS = [
("No. of people (Temp+Perm)", "People"),
("Air Conditioners", "ACs"),
("Air Coolers", "Coolers"),
("Refrigerators", "Fridges"),
("No. of rooms", "Rooms"),
("Property Area (Marla)", "Area (Marla)"),
]
def _fmt(value) -> str:
if value is None:
return "—"
if isinstance(value, float):
return f"{value:g}"
return str(value)
def profile_markdown(house_id: str) -> str:
house = BY_ID[house_id]
profile = house["profile"]
rows = "\n".join(
f"| {label} | {_fmt(profile.get(key))} |" for key, label in PROFILE_ROWS
)
return (
f"**{house_id} · {house['city']}**\n\n"
f"| Field | Value |\n|---|---|\n{rows}\n\n"
f"<sub>Origin {house['origin_time']} → target {house['target_time']}</sub>"
)
def defaults_for(house_id: str) -> list[float]:
latest = BY_ID[house_id]["defaults"]
return [float(latest[key]) for key, *_ in WEATHER_FIELDS]
def model_caption() -> str:
mae = INFO["test_mae"]
r2 = INFO["test_r2"]
coverage = INFO["interval_val_coverage"]
parts = [f"**Model:** {INFO['name']}"]
parts.append(f"test MAE {mae:.3f} kWh" if mae is not None else "test MAE n/a")
if r2 is not None:
parts.append(f"R² {r2:.3f}")
if coverage is not None:
parts.append(f"90% interval calibrated on {coverage:.0%} validation coverage")
parts.append(f"trained on {INFO['n_train_rows']:,} rows")
return " · ".join(parts)
def chart_for(result) -> plt.Figure:
frame = pd.DataFrame(result.history)
frame["time"] = pd.to_datetime(frame["time"])
fig, ax = plt.subplots(figsize=(8.2, 3.2))
ax.plot(frame["time"], frame["usage_kwh"], color="#1f77b4", label="last 72 h")
origin = frame["time"].iloc[-1]
ax.axvline(origin, color="grey", ls="--", lw=1, label="origin")
target = pd.Timestamp(result.target_time)
ax.errorbar(
[target],
[result.prediction_kwh],
yerr=[
[result.prediction_kwh - result.interval_low_kwh],
[result.interval_high_kwh - result.prediction_kwh],
],
fmt="o",
color="#d62728",
capsize=5,
label="forecast (90% interval)",
)
ax.set_ylabel("kWh")
ax.set_xlabel("datetime")
ax.legend(loc="upper left", fontsize=8)
ax.grid(alpha=0.3)
fig.autofmt_xdate()
fig.tight_layout()
return fig
def summary_for(result) -> str:
delta = result.delta_kwh
sign = "+" if delta >= 0 else "−"
return (
f"### {result.prediction_kwh:.3f} kWh\n"
f"**90% interval** {result.interval_low_kwh:.3f} – "
f"{result.interval_high_kwh:.3f} kWh \n"
f"**vs last hour** {sign}{abs(delta):.3f} kWh "
f"(last reading {result.last_hour_kwh:.3f} kWh) \n"
f"<sub>{result.house_id} · {result.city} · "
f"{result.origin_time} → {result.target_time}</sub>"
)
@spaces.GPU(duration=5) # no-op outside ZeroGPU; enables Space build
def gp_probe(x):
return x
def predict(house_id: str, *weather_values: float):
weather = {
key: float(value)
for (key, *_), value in zip(WEATHER_FIELDS, weather_values)
}
result = SERVICE.predict(house_id, weather)
return summary_for(result), chart_for(result)
def build_ui() -> gr.Blocks:
with gr.Blocks(title="BijliCast", theme=gr.themes.Soft()) as demo:
gr.Markdown(f"## ⚡ BijliCast\n*{model_caption()}*\n\n_Research/demo purposes only_")
with gr.Row():
with gr.Column(scale=5):
house = gr.Dropdown(
choices=HOUSE_IDS,
value=HOUSE_IDS[0],
label="Household",
)
profile = gr.Markdown(profile_markdown(HOUSE_IDS[0]))
sliders = [
gr.Slider(
minimum=low,
maximum=high,
value=float(BY_ID[HOUSE_IDS[0]]["defaults"][key]),
step=step,
label=label,
)
for key, label, low, high, step in WEATHER_FIELDS
]
button = gr.Button("⚡ Predict next hour", variant="primary")
with gr.Column(scale=6):
summary = gr.Markdown(
"_Pick a household and press **Predict next hour**._"
)
plot = gr.Plot(label="Last 72 hours + forecast")
gr.Markdown(
"<sub>Research/demo project. Not for billing or grid operations.</sub>"
)
house.change(
fn=lambda h: [profile_markdown(h), *defaults_for(h)],
inputs=house,
outputs=[profile, *sliders],
)
button.click(
fn=predict,
inputs=[house, *sliders],
outputs=[summary, plot],
)
return demo
demo = build_ui()
if __name__ == "__main__":
demo.launch()