Source code for quantdata.dysanos.implied.plotting

"""Publication figures for direct PCA-IV benchmark diagnostics."""

from __future__ import annotations

from collections.abc import Mapping
from pathlib import Path
from typing import Any

import numpy as np
from cdxcore.err import verify_inp

from .pca import LogIVPCA, PCAFitDiagnostics


def _save(figure: Any, output: str | Path) -> Path:
    import matplotlib.pyplot as plt

    path = Path(output)
    path.parent.mkdir(parents=True, exist_ok=True)
    figure.savefig(path, dpi=220)
    plt.close(figure)
    return path


[docs] def write_iv_eigensurface_figure( model: LogIVPCA, diagnostics: PCAFitDiagnostics, output: str | Path, ) -> Path: """Plot the first five one-standard-deviation IV eigensurfaces.""" import matplotlib.pyplot as plt surfaces = diagnostics.first_five_iv_eigensurfaces verify_inp( surfaces.ndim == 3 and surfaces.shape[1:] == model.surface_shape, lambda: f"invalid IV eigensurface geometry {surfaces.shape}", ) components = surfaces.shape[0] figure, axes = plt.subplots( components, 1, figsize=(8.0, 2.6 * components), constrained_layout=True, squeeze=False, ) magnitude = max(float(np.max(np.abs(surfaces))), np.finfo(float).eps) for component, axis in enumerate(axes[:, 0]): image = axis.pcolormesh( model.nms, np.broadcast_to(model.expiries[:, None], model.nms.shape), surfaces[component], shading="auto", cmap="coolwarm", vmin=-magnitude, vmax=magnitude, ) axis.axvline(0.0, color="black", linewidth=0.5, alpha=0.6) axis.set_title(f"log-IV level PC {component + 1}: one-score-std IV change") axis.set_ylabel("expiry (years)") figure.colorbar(image, ax=axis, label="implied-volatility change") axes[-1, 0].set_xlabel("normalized moneyness z") return _save(figure, output)
[docs] def write_pca_reconstruction_figure( model: LogIVPCA, diagnostics: PCAFitDiagnostics, output: str | Path, ) -> Path: """Plot explained variance and IV reconstruction error by expiry.""" import matplotlib.pyplot as plt figure, axes = plt.subplots(1, 2, figsize=(10.0, 3.8), constrained_layout=True) components = np.arange(1, model.num_components + 1) axes[0].bar(components, model.explained_variance_ratio, label="individual") axes[0].plot( components, model.cumulative_explained_variance_ratio, color="black", marker=".", label="cumulative", ) axes[0].set(xlabel="level component", ylabel="variance fraction") axes[0].legend() expiry = [item["expiry"] for item in diagnostics.errors_by_expiry] rmse = [item["iv_rmse"] for item in diagnostics.errors_by_expiry] maximum = [ item["iv_maximum_absolute_error"] for item in diagnostics.errors_by_expiry ] axes[1].plot(expiry, rmse, marker="o", label="RMSE") axes[1].plot(expiry, maximum, marker="o", label="maximum absolute") axes[1].set(xlabel="expiry (years)", ylabel="IV error") axes[1].legend() return _save(figure, output)
[docs] def write_distribution_fidelity_figure( invariant: Mapping[str, Any], weekly: Mapping[str, Any], output: str | Path, ) -> Path: """Plot matched invariant-level and weekly-increment distance metrics.""" import matplotlib.pyplot as plt names = ( "mean_rms_in_historical_std", "relative_covariance_frobenius", "sliced_wasserstein_in_historical_std", "coordinate_ks_mean", ) invariant_metrics = invariant["historical_comparison"] weekly_metrics = weekly[ "weekly_increment_vs_historical_weekly_increment" ] x = np.arange(len(names)) width = 0.38 figure, axis = plt.subplots(figsize=(9.0, 4.2), constrained_layout=True) axis.bar( x - width / 2, [invariant_metrics[name] for name in names], width, label="invariant levels", ) axis.bar( x + width / 2, [weekly_metrics[name] for name in names], width, label="weekly increments", ) axis.set_xticks(x, [name.replace("_", "\n") for name in names]) axis.set_ylabel("distance / discrepancy") axis.legend() return _save(figure, output)
[docs] def write_static_violation_figure( historical: Mapping[str, Any], generated: Mapping[str, Any], output: str | Path, ) -> Path: """Plot physical surface-violation rates across roundoff conventions.""" import matplotlib.pyplot as plt multipliers = [0, 1, 8, 32, 128] def fractions(report: Mapping[str, Any]) -> list[float]: values = report["by_roundoff_multiplier"] return [ values[str(multiplier)]["aggregate"][ "violating_surface_fraction" ] for multiplier in multipliers ] figure, axis = plt.subplots(figsize=(7.5, 4.0), constrained_layout=True) axis.plot(multipliers, fractions(historical), marker="o", label="historical PCA") axis.plot(multipliers, fractions(generated), marker="o", label="generated PCA") axis.set( xlabel="float32 roundoff multiplier", ylabel="surfaces with a violation", yscale="symlog", ) axis.legend() return _save(figure, output)
[docs] def write_minority_tail_figure( reports: Mapping[str, Mapping[str, Any]], output: str | Path ) -> Path: """Plot the twenty smallest physical minority-sign probabilities.""" import matplotlib.pyplot as plt figure, axis = plt.subplots(figsize=(8.0, 4.2), constrained_layout=True) for split, report in reports.items(): values = np.asarray( report["per_instrument"]["weighted_minority_probability"], dtype=np.float64, )[1:] axis.plot( np.arange(1, min(20, values.size) + 1), np.sort(values)[:20], marker="o", label=split, ) axis.set( xlabel="rank among option instruments", ylabel="physical minority-sign probability", ) axis.legend() return _save(figure, output)
[docs] def write_local_candidate_figure( report: Mapping[str, Any], output: str | Path ) -> Path: """Plot ordinary local one-sided candidate states by split/direction.""" import matplotlib.pyplot as plt splits = ("training", "validation") directions = ("long_only", "short_only", "both_directions") values = np.asarray( [ [ report[split]["zero_adverse"][ "states_by_direction_exclusivity" ][direction] for direction in directions ] for split in splits ], dtype=np.float64, ) x = np.arange(len(splits)) width = 0.25 figure, axis = plt.subplots(figsize=(7.0, 4.0), constrained_layout=True) for index, direction in enumerate(directions): axis.bar( x + (index - 1) * width, values[:, index], width, label=direction.replace("_", " "), ) axis.set_xticks(x, splits) axis.set_ylabel("query states with candidates") axis.legend() return _save(figure, output)
__all__ = [ "write_distribution_fidelity_figure", "write_iv_eigensurface_figure", "write_local_candidate_figure", "write_minority_tail_figure", "write_pca_reconstruction_figure", "write_static_violation_figure", ]