plotting

name	plotting
description	Use this skill any time a forecast chart, visualization, or figure is created. This includes: plotting single forecasts, grid comparisons, counterfactual scenario fans, backtest result charts, or any matplotlib figure using Migas-1.5 outputs. Trigger whenever the user mentions plotting, charting, visualizing, or saving forecast figures.

Plotting Migas-1.5 Forecasts

Always call apply_migas_style() once before any plotting to get consistent, release-quality styling.

Setup

from migaseval.plotting_utils import (
    COLORS,
    apply_migas_style,
    plot_forecast_single,
    plot_forecast_grid,
    _draw_forecast_region,
)

apply_migas_style()  # call once — sets rcParams for clean, modern look

Color Palette

Use the COLORS dict for consistent styling across all plots:

Key	Hex	Use
"historical"	#5D6D7E	Context window (slate gray)
"ground_truth"	#1B2631	Actual future values (charcoal)
"Migas-1.5"	#F28C28	Migas-1.5 forecast (vibrant teal/orange — hero color)
"Chronos-2"	#4A90D9	Chronos baseline (steel blue)
"forecast_region"	#FDF6EC	Background shading for forecast zone
"forecast_vline"	#ABB2B9	Vertical divider at forecast boundary

For counterfactual scenarios, the notebooks use:

• Bullish: #2EAD6D (green)
• Bearish: #C0392B (red)
• Scenario fill: #9B8EC4 with alpha=0.08

Single Forecast Plot

plot_forecast_single creates a one-window figure with optional ground truth, metrics badge, and text summary panel below the chart.

fig, ax = plot_forecast_single(
    history,                          # np.ndarray, shape (context_len,)
    gt_vals,                          # np.ndarray, shape (pred_len,) or None
    {"Chronos-2": chronos_fc, "Migas-1.5": migas_fc},  # dict of forecasts
    context_len,                      # int — SEQ_LEN
    pred_len,                         # int — PRED_LEN
    title="Chronos-2 vs. Migas-1.5",
    figsize=(11, 4),
    show_metrics=True,                # adds MAPE/MAE badge when gt is provided
    text_summary=summary,             # optional — renders summary below the plot
    timestamps=full["t"].values,      # optional — uses dates on x-axis instead of step indices
)
plt.show()

Key behaviors:

• When timestamps is provided (length = context_len + pred_len), the x-axis shows dates
• When text_summary is provided, a styled text panel is added below the chart
• When show_metrics=True and ground truth exists, a MAPE/MAE badge appears above the title
• Migas-1.5 is always drawn on top (highest z-order)

Grid of Forecast Windows

plot_forecast_grid creates a multi-subplot figure for comparing multiple windows (e.g. from a backtest or batch run).

fig, axes = plot_forecast_grid(
    history_2d,                       # np.ndarray, shape (n_samples, context_len)
    gt_2d,                            # np.ndarray, shape (n_samples, pred_len) or None
    {"Migas-1.5": migas_2d, "Chronos-2": chronos_2d},  # dict of (n_samples, pred_len)
    context_len,
    pred_len,
    sample_indices=[0, 1, 2, 3],      # which rows to plot
    titles=["Window 1", "Window 2", "Window 3", "Window 4"],  # optional
    max_cols=3,                        # max columns in grid
    show_metrics=True,
)
plt.show()

Forecast Region Shading

_draw_forecast_region adds a warm-tinted background band and vertical divider to mark the forecast zone. Used internally by the plot functions, but can be called directly for custom plots:

_draw_forecast_region(
    ax, context_len, pred_len,
    boundary=t_pred[0],       # datetime or int — left edge of forecast zone
    boundary_end=t_pred[-1],  # datetime or int — right edge
)

Counterfactual Scenario Fan (Custom Plot)

For the scenario fan plot (original + bullish + bearish), build it manually since it needs custom colors and fill. Pattern from the notebooks:

import matplotlib.dates as mdates

fig, ax = plt.subplots(figsize=(11, 5))
_draw_forecast_region(ax, CONTEXT_LEN, PRED_LEN, boundary=t_pred[0], boundary_end=t_pred[-1])

# Historical
ax.plot(t_ctx, context_vals, color=COLORS["historical"], lw=2.0, label="Historical")

# Ground truth (if available)
ax.plot(t_pred, np.concatenate([[last_val], gt_vals]),
        color=COLORS["ground_truth"], lw=2.2, label="Ground Truth")

# Original forecast
ax.plot(t_pred, np.concatenate([[last_val], fc_original]),
        color=COLORS["Migas-1.5"], lw=2.0, ls="--", alpha=0.85, label="Original (Migas-1.5)")

# Bullish / Bearish
ax.plot(t_pred, np.concatenate([[last_val], fc_bullish]),
        color="#2EAD6D", lw=2.4, label="Bullish scenario")
ax.plot(t_pred, np.concatenate([[last_val], fc_bearish]),
        color="#C0392B", lw=2.4, label="Bearish scenario")

# Scenario range fill
ax.fill_between(t_pred,
    np.concatenate([[last_val], fc_bearish]),
    np.concatenate([[last_val], fc_bullish]),
    alpha=0.08, color="#9B8EC4", label="Scenario range")

ax.xaxis.set_major_formatter(mdates.ConciseDateFormatter(ax.xaxis.get_major_locator()))
ax.legend(fontsize=8, handlelength=1.6)
fig.tight_layout(pad=1.2)
plt.show()

Important Notes

• Always prepend last_val (the last context value) to forecast arrays when plotting prediction lines — this connects the forecast to the historical line visually: np.concatenate([[last_val], forecast])
• t_pred should have length pred_len + 1 to match the prepended arrays
• For date x-axes, rotate labels: lbl.set_rotation(35); lbl.set_ha("right")
• Denormalization is built-in: pass history_mean and history_std to automatically rescale values back to original units