app.py

import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import gradio as gr
from darts import TimeSeries
from darts.models import TFTModel, NBEATSModel
from darts.dataprocessing.transformers import Scaler
from sklearn.preprocessing import LabelEncoder
import numpy as np
import io
import os

# ----------------------------
# SAFE DATASET LOADER
# ----------------------------
def load_dataset(path="dataset.csv", url=None):
    try:
        # Try UTF-8
        return pd.read_csv(path, encoding="utf-8")
    except UnicodeDecodeError:
        try:
            # Fallback to Latin1
            return pd.read_csv(path, encoding="latin1")
        except Exception as e:
            if url:
                # If file missing, try URL
                try:
                    return pd.read_csv(url, encoding="utf-8")
                except UnicodeDecodeError:
                    return pd.read_csv(url, encoding="latin1")
            else:
                raise e

# Path or fallback URL
url = "https://raw.githubusercontent.com/yourusername/yourrepo/main/dataset.csv"
if os.path.exists("dataset.csv"):
    df = load_dataset("dataset.csv")
else:
    df = load_dataset(url=url)

# ----------------------------
# Preprocessing
# ----------------------------
df['datetime'] = pd.to_datetime(df['datetime'])
df = df.sort_values("datetime")

# Encode weather icons
encoder = LabelEncoder()
if "icon" in df.columns:
    df['icon_encoded'] = encoder.fit_transform(df['icon'])

# Create timeseries
series = TimeSeries.from_dataframe(df, "datetime", "pv_output_kWh")
scaler = Scaler()
series_scaled = scaler.fit_transform(series)

# Pre-trained or fallback model
try:
    model = TFTModel.load_from_checkpoint("tft_pretrained", work_dir="./")
except Exception:
    model = NBEATSModel(input_chunk_length=30, output_chunk_length=7, n_epochs=10)
    model.fit(series_scaled)

# ----------------------------
# EDA FUNCTIONS
# ----------------------------
def eda_summary():
    buf = io.StringIO()
    df.describe().to_string(buf)
    return buf.getvalue()

def eda_histogram(column):
    plt.figure(figsize=(6,4))
    sns.histplot(df[column], kde=True, bins=20)
    plt.title(f"Distribution of {column}")
    plt.tight_layout()
    return plt.gcf()

def eda_correlation():
    plt.figure(figsize=(8,6))
    sns.heatmap(df.corr(), annot=True, cmap="coolwarm", fmt=".2f")
    plt.title("Correlation Heatmap")
    plt.tight_layout()
    return plt.gcf()

def eda_timeseries():
    plt.figure(figsize=(10,4))
    plt.plot(df["datetime"], df["pv_output_kWh"], label="PV Output (kWh)")
    plt.title("Time-Series Trend of PV Output")
    plt.xlabel("Date")
    plt.ylabel("PV Output (kWh)")
    plt.legend()
    plt.tight_layout()
    return plt.gcf()

# ----------------------------
# FORECAST FUNCTION
# ----------------------------
def forecast_pv(horizon, weather_condition):
    horizon_map = {"24 Hours": 24, "3 Days": 72, "7 Days": 168, "14 Days": 336}
    steps = horizon_map[horizon]

    forecast = model.predict(steps)
    forecast = scaler.inverse_transform(forecast)

    # Weather impact adjustment
    adjustment = {
        "Clear": 1.0,
        "Partly Cloudy": 0.85,
        "Cloudy": 0.65,
        "Fog": 0.55,
        "Smoke/Dust": 0.6,
        "Winter": 0.7,
        "Rain": 0.5
    }
    adj_factor = adjustment.get(weather_condition, 1.0)
    forecast_adj = forecast * adj_factor

    # Plot
    plt.figure(figsize=(10,4))
    series[-7*24:].plot(label="History")  # last week history
    forecast.plot(label="Forecast (Base)")
    forecast_adj.plot(label=f"Forecast (Adjusted: {weather_condition})")
    plt.legend()
    plt.title(f"PV Forecast for {horizon}")
    plt.tight_layout()

    # Peak info
    peak_time = forecast_adj.time_index[np.argmax(forecast_adj.values())]
    peak_val = np.max(forecast_adj.values())
    peak_info = f"🔺 Peak PV Output: {round(peak_val,2)} kWh at {peak_time}"

    return plt.gcf(), peak_info

# ----------------------------
# GRADIO DASHBOARD
# ----------------------------
eda_tab = gr.TabbedInterface(
    [
        gr.Interface(fn=eda_summary, inputs=[], outputs="text", title="Summary Stats"),
        gr.Interface(fn=eda_histogram, inputs=gr.Dropdown(df.columns, label="Select Column"), outputs="plot", title="Histogram"),
        gr.Interface(fn=eda_correlation, inputs=[], outputs="plot", title="Correlation Heatmap"),
        gr.Interface(fn=eda_timeseries, inputs=[], outputs="plot", title="Time Series Trend")
    ],
    tab_names=["Summary", "Histogram", "Correlation", "Time Series"]
)

forecast_tab = gr.Interface(
    fn=forecast_pv,
    inputs=[
        gr.Radio(["24 Hours", "3 Days", "7 Days", "14 Days"], label="Select Forecast Horizon"),
        gr.Dropdown(["Clear","Partly Cloudy","Cloudy","Fog","Smoke/Dust","Winter","Rain"], label="Weather Condition")
    ],
    outputs=[
        gr.Plot(label="Forecast Plot"),
        gr.Textbox(label="Peak Info")
    ],
    title="PV Forecasting"
)

app = gr.TabbedInterface([eda_tab, forecast_tab], tab_names=["EDA Dashboard", "Forecasting"])

if __name__ == "__main__":
    app.launch()