source/app.py

from __future__ import annotations

import json
from pathlib import Path

import gradio as gr
import plotly.graph_objects as go
import torch
from data import multiscale_batch
from model import ClockworkRNN, MatchedGRU, PlainRNN
from safetensors.torch import load_file

ARTIFACT_DIR = Path(__file__).resolve().parent / "artifacts" / "clockwork-rnn-pocket"
REPORT = json.loads((ARTIFACT_DIR / "evaluation.json").read_text(encoding="utf-8"))
MODELS = {
    "Clockwork RNN": (ClockworkRNN(), "clockwork_rnn"),
    "Plain RNN": (PlainRNN(), "plain_rnn"),
    "Matched GRU": (MatchedGRU(), "matched_gru"),
}
for model, key in MODELS.values():
    model.load_state_dict(load_file(ARTIFACT_DIR / f"{key}.safetensors"))
    model.eval()


@torch.inference_mode()
def compare(seed: int, length: int) -> tuple[go.Figure, dict]:
    sequence = multiscale_batch(1, int(length) + 1, int(seed))
    target = sequence[0, 1:, 0]
    figure = go.Figure()
    figure.add_trace(go.Scatter(y=target, name="Target", line={"width": 4}))
    metrics = {}
    for label, (model, key) in MODELS.items():
        prediction = model(sequence[:, :-1])[0, :, 0]
        figure.add_trace(go.Scatter(y=prediction, name=label))
        metrics[label] = {
            "live_rmse": float((prediction - target).square().mean().sqrt()),
            "verified_length_256_rmse": REPORT["results"][key][
                "length_256_zero_shot"
            ]["rmse"],
        }
    figure.update_layout(
        template="plotly_dark",
        title="Teacher-forced next-step multiscale forecast",
        xaxis_title="Time",
        yaxis_title="Signal",
    )
    return figure, metrics


with gr.Blocks(title="Clockwork RNN Pocket") as demo:
    gr.Markdown(
        "# Clockwork RNN Pocket\n"
        "Compare periodic hidden-state updates with parameter-matched recurrent "
        "controls on a multiscale signal."
    )
    with gr.Row():
        seed = gr.Slider(0, 100_000, value=2099, step=1, label="Signal seed")
        length = gr.Slider(32, 256, value=128, step=16, label="Sequence length")
    initial = compare(2099, 128)
    chart = gr.Plot(value=initial[0])
    metrics = gr.JSON(value=initial[1])
    button = gr.Button("Run recurrent retest", variant="primary")
    button.click(compare, inputs=[seed, length], outputs=[chart, metrics])


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