import numpy as np
import pandas as pd
import gradio as gr
import tensorflow as tf

MODEL_PATH = "activity_model.keras"
CLASS_NAMES = ["Stationary", "Walking", "Running"]
EXPECTED_STEPS = 100

model = tf.keras.models.load_model(MODEL_PATH)

def read_signal(file_path):
    if file_path is None:
        raise gr.Error("Please upload a CSV file.")
    
    df = pd.read_csv(file_path)
    
    # Accept either a column named sensor_value or the first numeric column.
    if "sensor_value" in df.columns:
        values = df["sensor_value"].to_numpy(dtype=np.float32)
    else:
        numeric = df.select_dtypes(include=["number"])
        if numeric.shape[1] == 0:
            raise gr.Error("CSV must contain a numeric sensor column.")
        values = numeric.iloc[:, 0].to_numpy(dtype=np.float32)
        
    if len(values) != EXPECTED_STEPS:
        raise gr.Error(f"Expected exactly {EXPECTED_STEPS} sensor readings, found {len(values)}.")
        
    return values

def predict_activity(file_path):
    values = read_signal(file_path)
    x = values.reshape(1, EXPECTED_STEPS, 1)
    probs = model.predict(x, verbose=0)[0]
    return {name: float(prob) for name, prob in zip(CLASS_NAMES, probs)}

with gr.Blocks() as demo:
    gr.Markdown("# Wearable Activity Classifier")
    gr.Markdown("Upload a CSV containing exactly 100 sensor readings. The student-trained model predicts Stationary, Walking, or Running.")
    file_input = gr.File(label="Sensor CSV", file_types=[".csv"], type="filepath")
    output = gr.Label(label="Prediction", num_top_classes=3)
    button = gr.Button("Predict activity", variant="primary")
    button.click(predict_activity, inputs=file_input, outputs=output)
    
    gr.Examples(
        examples=[["sample_stationary.csv"], ["sample_walking.csv"], ["sample_running.csv"]],
        inputs=file_input,
        label="Sample files"
    )

demo.launch()