app.py

import io
import os
from pathlib import Path
from typing import Dict, List, Optional, Tuple

FEATURE_NAMES = [
    "temperature_f",
    "humidity_percent",
    "wind_mph",
    "hour_of_day",
    "is_weekend",
]


CACHE_ROOT = Path(__file__).with_name(".cache")
CACHE_ROOT.mkdir(exist_ok=True)
matplotlib_cache = CACHE_ROOT / "matplotlib"
matplotlib_cache.mkdir(parents=True, exist_ok=True)
os.environ.setdefault("MPLCONFIGDIR", str(matplotlib_cache))
os.environ.setdefault("XDG_CACHE_HOME", str(CACHE_ROOT))

import gradio as gr
import matplotlib

matplotlib.use("Agg")
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from PIL import Image
from sklearn.metrics import accuracy_score, classification_report
from sklearn.model_selection import train_test_split
from sklearn.tree import DecisionTreeClassifier, plot_tree


CUSTOM_CSS = """
#predict-button {
    background: linear-gradient(135deg, #ce1126, #8b0000);
    color: white;
    font-weight: 700;
    font-size: 1.05rem;
    border: none;
    box-shadow: 0 6px 18px rgba(139, 0, 0, 0.35);
}
#predict-button:hover {
    transform: translateY(-1px);
    box-shadow: 0 10px 24px rgba(139, 0, 0, 0.45);
}
.prediction-card {
    border: 3px solid #ce1126;
    border-radius: 18px;
    padding: 1.5rem;
    background: #fff4f4;
    text-align: center;
    box-shadow: 0 12px 30px rgba(206, 17, 38, 0.15);
}
.prediction-card .location {
    font-size: 2rem;
    font-weight: 800;
    color: #7b0011;
    letter-spacing: 0.5px;
}
.prediction-card .location span {
    text-transform: uppercase;
}
.prediction-card .confidence {
    margin-top: 0.75rem;
    font-size: 1.05rem;
    color: #333;
}
.prediction-card .secondary {
    margin-top: 0.25rem;
    font-size: 0.95rem;
    color: #555;
}
.path-list {
    list-style: none;
    padding: 0;
    margin: 0;
    display: flex;
    flex-direction: column;
    gap: 0.75rem;
}
.path-list li {
    background: #f8f9fa;
    border-left: 5px solid #ce1126;
    padding: 0.75rem 1rem;
    border-radius: 10px;
    box-shadow: inset 0 0 0 1px rgba(0, 0, 0, 0.05);
}
.path-list li .headline {
    font-weight: 700;
    color: #7b0011;
    margin-bottom: 0.2rem;
}
.path-list li .meta {
    color: #333;
    font-size: 0.95rem;
}
.path-list li.leaf {
    border-left-color: #1b5e20;
    background: #e8f5e9;
}
.path-list li.leaf .headline {
    color: #1b5e20;
}
"""


def build_dataset(n_samples: int = 200, seed: int = 42) -> pd.DataFrame:
    rng = np.random.default_rng(seed)
    data = pd.DataFrame(
        {
            "temperature_f": rng.integers(60, 115, n_samples),
            "humidity_percent": rng.integers(10, 40, n_samples),
            "wind_mph": rng.integers(0, 25, n_samples),
            "hour_of_day": rng.integers(8, 22, n_samples),
            "is_weekend": rng.integers(0, 2, n_samples),
        }
    )

    labels: list[str] = []
    for idx in range(n_samples):
        temp = data.at[idx, "temperature_f"]
        wind = data.at[idx, "wind_mph"]
        hour = data.at[idx, "hour_of_day"]
        if temp < 85 and wind < 15 and 8 <= hour <= 18:
            labels.append("Outdoors")
        elif temp > 105:
            labels.append("Library")
        elif wind > 20:
            labels.append("Library")
        elif hour > 19:
            labels.append("Library")
        else:
            labels.append(rng.choice(["Library", "Outdoors"], p=[0.6, 0.4]))

    data["study_location"] = labels
    return data


def train_model(data: pd.DataFrame) -> Tuple[DecisionTreeClassifier, Dict[str, float]]:
    X = data[FEATURE_NAMES]
    y = data["study_location"]
    X_train, X_test, y_train, y_test = train_test_split(
        X, y, test_size=0.2, random_state=42
    )
    clf = DecisionTreeClassifier(
        max_depth=3,
        min_samples_split=10,
        random_state=42,
    )
    clf.fit(X_train, y_train)

    y_pred_train = clf.predict(X_train)
    y_pred_test = clf.predict(X_test)
    metrics = {
        "train_accuracy": accuracy_score(y_train, y_pred_train),
        "test_accuracy": accuracy_score(y_test, y_pred_test),
    }
    return clf, metrics


def describe_path(
    model: DecisionTreeClassifier, sample: np.ndarray
) -> Tuple[List[Dict[str, object]], List[int]]:
    tree = model.tree_
    node_indicator = model.decision_path(sample.reshape(1, -1))
    leaf_id = model.apply(sample.reshape(1, -1))[0]
    start, end = node_indicator.indptr[:2]
    node_index = node_indicator.indices[start:end]

    steps: List[Dict[str, object]] = []
    path_nodes = list(node_index) + [leaf_id]
    for node_position, node_id in enumerate(node_index, start=1):
        feature_index = int(tree.feature[node_id])
        threshold = float(tree.threshold[node_id])
        feature_name = FEATURE_NAMES[feature_index]
        feature_value = float(sample[feature_index])
        go_left = feature_value <= threshold
        direction = "≤" if go_left else ">"
        next_node = int(tree.children_left[node_id] if go_left else tree.children_right[node_id])
        steps.append(
            {
                "step": node_position,
                "node_id": int(node_id),
                "next_node": next_node,
                "feature": feature_name,
                "feature_label": feature_name.replace("_", " ").title(),
                "threshold": threshold,
                "value": feature_value,
                "direction": direction,
                "decision": "left" if go_left else "right",
            }
        )

    leaf_samples = int(tree.n_node_samples[leaf_id])
    confidences = model.predict_proba(sample.reshape(1, -1))[0]
    class_idx = int(confidences.argmax())
    class_name = model.classes_[class_idx]
    steps.append(
        {
            "step": len(node_index) + 1,
            "node_id": int(leaf_id),
            "leaf": True,
            "prediction": class_name,
            "confidence": float(confidences[class_idx]),
            "samples": leaf_samples,
        }
    )
    return steps, path_nodes


def render_tree_image(
    model: DecisionTreeClassifier,
    highlighted_nodes: Optional[List[int]] = None,
    leaf_id: Optional[int] = None,
) -> Image.Image:
    highlighted = set(highlighted_nodes or [])
    fig, ax = plt.subplots(figsize=(10, 6))
    texts = plot_tree(
        model,
        feature_names=[name.replace("_", " ").title() for name in FEATURE_NAMES],
        class_names=model.classes_,
        filled=False,
        rounded=True,
        node_ids=True,
        fontsize=9,
        ax=ax,
    )
    fig.tight_layout()
    for text in texts:
        label = text.get_text()
        node_id = None
        for line in label.split("\n"):
            if line.startswith("node id ="):
                try:
                    node_id = int(line.split("=")[1].strip())
                except ValueError:
                    node_id = None
                break
        if node_id is None:
            continue
        if node_id in highlighted:
            is_leaf = leaf_id is not None and node_id == leaf_id
            text.set_bbox(
                dict(
                    boxstyle="round,pad=0.45",
                    facecolor="#e8f5e9" if is_leaf else "#ffe5e9",
                    edgecolor="#2e7d32" if is_leaf else "#ce1126",
                    linewidth=2.5,
                )
            )
            text.set_color("#1b5e20" if is_leaf else "#7b0011")
        else:
            text.set_bbox(
                dict(
                    boxstyle="round,pad=0.45",
                    facecolor="#f5f7fa",
                    edgecolor="#c3cfe2",
                    linewidth=1.2,
                )
            )
            text.set_color("#2e2e2e")

    buffer = io.BytesIO()
    fig.savefig(buffer, format="png", dpi=200, bbox_inches="tight")
    plt.close(fig)
    buffer.seek(0)
    return Image.open(buffer)


def load_html_snippet() -> str:
    html_path = Path(__file__).with_name("interactive_decision_tree.html")
    if html_path.exists():
        return html_path.read_text(encoding="utf-8")
    return ""


DATAFRAME = build_dataset()
MODEL, METRICS = train_model(DATAFRAME)
HTML_SNIPPET = load_html_snippet()
CLASS_REPORT = classification_report(
    DATAFRAME["study_location"],
    MODEL.predict(DATAFRAME[FEATURE_NAMES]),
    target_names=MODEL.classes_,
    zero_division=0,
)
DEFAULT_TREE_IMAGE = render_tree_image(MODEL)


def predict_study_location(
    temperature: int,
    humidity: int,
    wind: int,
    hour: int,
    weekend: bool,
) -> Tuple[str, str, str, Image.Image]:
    sample = np.array(
        [
            temperature,
            humidity,
            wind,
            hour,
            1 if weekend else 0,
        ],
        dtype=float,
    )
    probabilities = MODEL.predict_proba(sample.reshape(1, -1))[0]
    top_index = probabilities.argmax()
    label = MODEL.classes_[top_index]
    confidence = probabilities[top_index]
    secondary_index = (
        probabilities.argsort()[::-1][1]
        if len(probabilities) > 1
        else top_index
    )
    secondary_confidence = probabilities[secondary_index]
    secondary_label = MODEL.classes_[secondary_index]
    step_details, path_nodes = describe_path(MODEL, sample)
    leaf_id = path_nodes[-1] if path_nodes else None
    path_html_items: list[str] = []
    for detail in step_details:
        if detail.get("leaf"):
            path_html_items.append(
                f"""
                <li class="leaf">
                    <div class="headline">Leaf: predict {detail['prediction']}</div>
                    <div class="meta">Confidence {detail['confidence']:.0%} · Support {detail['samples']} samples</div>
                </li>
                """.strip()
            )
        else:
            threshold = detail["threshold"]
            value = detail["value"]
            direction = detail["direction"]
            feature_label = detail["feature_label"]
            decision = detail["decision"]
            next_node = detail["next_node"]
            path_html_items.append(
                f"""
                <li>
                    <div class="headline">
                        Step {detail['step']}: {feature_label} {direction} {threshold:.1f}
                    </div>
                    <div class="meta">
                        Observed value {value:.1f} → take {decision} branch (node {next_node})
                    </div>
                </li>
                """.strip()
            )

    path_html = "<ol class='path-list'>" + "\n".join(path_html_items) + "</ol>"
    prediction_html = f"""
    <div class="prediction-card">
        <div class="location">Study at the <span>{label}</span></div>
        <div class="confidence">Confidence: {confidence:.1%}</div>
        <div class="secondary">Next best: {secondary_label} ({secondary_confidence:.1%})</div>
    </div>
    """.strip()
    confidence_text = (
        f"Primary recommendation: **{label}** (`{confidence:.1%}`) · "
        f"Alternate: **{secondary_label}** (`{secondary_confidence:.1%}`)"
    )
    highlighted_image = render_tree_image(
        MODEL,
        highlighted_nodes=path_nodes,
        leaf_id=leaf_id,
    )
    return prediction_html, confidence_text, path_html, highlighted_image


with gr.Blocks(
    title="UNLV Study Location Predictor",
    css=CUSTOM_CSS,
) as demo:
    gr.Markdown(
        """
        # UNLV Study Location Predictor
        Adjust the sliders to mirror the current Las Vegas weather and the decision tree
        will suggest the best study location. All numbers come from a synthetic dataset
        tailored for classroom walkthroughs.
        """
    )

    with gr.Row():
        with gr.Column():
            temperature = gr.Slider(
                minimum=60,
                maximum=115,
                value=85,
                step=1,
                label="Temperature (°F)",
                info="Normal daytime range in Las Vegas",
            )
            humidity = gr.Slider(
                minimum=10,
                maximum=40,
                value=20,
                step=1,
                label="Humidity (%)",
            )
            wind = gr.Slider(
                minimum=0,
                maximum=25,
                value=10,
                step=1,
                label="Wind Speed (mph)",
            )
            hour = gr.Slider(
                minimum=8,
                maximum=22,
                value=14,
                step=1,
                label="Hour of Day (24h)",
            )
            weekend = gr.Checkbox(
                label="Is it the weekend?",
                value=False,
            )
            run_button = gr.Button(
                "Predict Study Location",
                elem_id="predict-button",
            )

        with gr.Column():
            prediction_box = gr.HTML()
            confidence_box = gr.Markdown()
            path_box = gr.HTML(
                "<p style='color:#666;font-style:italic;'>Run a prediction to walk through each decision rule.</p>"
            )

    with gr.Accordion("Explore the Decision Tree", open=False):
        tree_image = gr.Image(
            value=DEFAULT_TREE_IMAGE,
            image_mode="RGB",
            show_label=False,
        )
        gr.Markdown(
            f"Train accuracy: `{METRICS['train_accuracy']:.1%}` | "
            f"Test accuracy: `{METRICS['test_accuracy']:.1%}`"
        )
        gr.Markdown(
            """
            **Legend:**  
            • *gini* – how mixed the classes are (0 = pure, higher = more mixed)  
            • *samples* – number of training rows that reached the node  
            • *class* – majority label assigned when the tree predicts at that node
            """,
        )

    with gr.Accordion("Inspect the Synthetic Dataset", open=False):
        gr.DataFrame(
            value=DATAFRAME.head(20),
            wrap=True,
            label="Sample of the training data (20 rows)",
            interactive=False,
        )
        gr.Markdown(
            "Class balance and precision/recall on the full dataset:"
            f"\n```\n{CLASS_REPORT}\n```"
        )

    if HTML_SNIPPET:
        with gr.Accordion("HTML Prototype (original demo)", open=False):
            gr.HTML(HTML_SNIPPET)

    run_button.click(
        predict_study_location,
        inputs=[temperature, humidity, wind, hour, weekend],
        outputs=[prediction_box, confidence_box, path_box, tree_image],
    )


if __name__ == "__main__":
    queued_app = demo.queue()
    is_hf_space = bool(os.environ.get("SPACE_ID"))
    default_port = int(
        os.environ.get(
            "PORT",
            os.environ.get("GRADIO_SERVER_PORT", "7860"),
        )
    )
    launch_kwargs = {
        "server_name": os.environ.get("GRADIO_SERVER_NAME", "0.0.0.0"),
        "server_port": default_port,
        "show_error": True,
        "share": os.environ.get("GRADIO_SHARE", "").lower() == "true",
    }
    if not is_hf_space:
        launch_kwargs["prevent_thread_lock"] = True

    try:
        queued_app.launch(**launch_kwargs)
    except OSError:
        fallback_port = int(
            os.environ.get(
                "GRADIO_FALLBACK_PORT",
                str(default_port + 1111),
            )
        )
        if fallback_port == launch_kwargs["server_port"]:
            raise
        launch_kwargs["server_port"] = fallback_port
        queued_app.launch(**launch_kwargs)