FASTWIND_emulator/parallel_plot_thesis.py at main · API-AXIOM/FASTWIND_emulator · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
#!/usr/bin/env python3
"""Create thesis-ready parallel-coordinate plots from exported 5-par metadata."""

import argparse
import json
from pathlib import Path

import matplotlib as mpl
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import plotly.graph_objects as go


DISPLAY = [
    ("activation_code", "Activation"),
    ("latent_dim", "Latent\ndimension"),
    ("fourier_modes", "Fourier\nmodes"),
    ("dropout", "Dropout"),
    ("learning_rate_log", "Learning rate"),
    ("batch_size", "Batch size"),
    ("max_epochs", "Maximum\nepochs"),
    ("early_stop_count", "Early-stopped\nlines"),
    ("call_time_log_ms", "Call time\n(ms, log scale)"),
    ("branch_depth", "Branch\ndepth"),
    ("trunk_depth", "Trunk\ndepth"),
    ("branch_width_max", "Maximum\nbranch width"),
    ("trunk_width_max", "Maximum\ntrunk width"),
    ("val_loss_log", "Validation loss\n(log scale)"),
    ("mse_log", "Test MSE\n(log scale)"),
    ("mare_log", "Test MARE\n(log scale)"),
]

AXIS_PAD = 0.04


def load_data(path):
    rows = json.loads(Path(path).read_text())["architecture_level"]
    df = pd.DataFrame(rows)
    df = df[df["activation"].isin(["relu", "gelu"])].copy()
    numeric = [x[0] for x in DISPLAY if x[0] not in {"activation_code", "learning_rate_log", "call_time_log_ms", "val_loss_log", "mse_log", "mare_log"}]
    numeric += ["learning_rate", "total_emulator_call_time_sec", "val_loss", "mse", "mare"]
    for col in numeric:
        df[col] = pd.to_numeric(df[col], errors="coerce")
    df["activation_code"] = df["activation"].map({"relu": 0.0, "gelu": 1.0})
    df["learning_rate_log"] = np.log10(df["learning_rate"])
    df["call_time_log_ms"] = np.log10(1000.0 * df["total_emulator_call_time_sec"])
    for metric in ("val_loss", "mse", "mare"):
        df[f"{metric}_log"] = np.log10(df[metric])
    df = df.dropna(subset=["val_loss", "mse", "mare"]).reset_index(drop=True)
    return df


def tick_spec(key, values):
    values = np.asarray(values, dtype=float)
    values = values[np.isfinite(values)]
    unique = np.unique(values)
    if key == "activation_code":
        return [0, 1], ["ReLU", "GELU"]
    if key == "learning_rate_log":
        return unique, [f"{10**v:.0e}" for v in unique]
    if key == "call_time_log_ms":
        raw = np.array([5, 10, 20, 50, 100, 200], dtype=float)
        keep = raw[(raw >= 10**values.min() * 0.85) & (raw <= 10**values.max() * 1.15)]
        return np.log10(keep), [f"{v:g}" for v in keep]
    if key.endswith("_log"):
        raw = np.geomspace(10**values.min(), 10**values.max(), 4)
        return np.log10(raw), [f"{v:.1e}" for v in raw]
    if len(unique) <= 8:
        return unique, [f"{v:g}" for v in unique]
    ticks = np.linspace(values.min(), values.max(), 5)
    return ticks, [f"{v:g}" for v in ticks]


def normalized(df):
    vals = df[[x[0] for x in DISPLAY]].to_numpy(float)
    lows = np.nanmin(vals, axis=0)
    highs = np.nanmax(vals, axis=0)
    spans = np.where(highs > lows, highs - lows, 1.0)
    return (vals - lows) / spans, lows, spans


def draw_static(df, output, highlighted=False):
    norm, lows, spans = normalized(df)
    best = int(df["mare"].idxmin())
    cmap = mpl.colormaps["viridis"]
    metric_norm = mpl.colors.Normalize(df["mare_log"].min(), df["mare_log"].max())
    fig, host = plt.subplots(figsize=(18.5, 8.2), constrained_layout=False)
    x = np.arange(len(DISPLAY))
    if highlighted:
        for row in norm:
            host.plot(x, row, color="#c9c9c9", lw=0.55, alpha=0.32, zorder=1)
        host.plot(x, norm[best], color=cmap(metric_norm(df.loc[best, "mare_log"])), lw=3.2, alpha=1, zorder=4)
    else:
        order = np.argsort(df["mare"].to_numpy())[::-1]
        for idx in order:
            host.plot(x, norm[idx], color=cmap(metric_norm(df.loc[idx, "mare_log"])), lw=0.8, alpha=0.48, zorder=2)
    host.set_xlim(0, len(DISPLAY) - 1)
    host.set_ylim(-AXIS_PAD, 1 + AXIS_PAD)
    host.axis("off")
    for j, (key, label) in enumerate(DISPLAY):
        ax = host.twinx()
        ax.spines["right"].set_position(("axes", j / (len(DISPLAY) - 1)))
        ax.spines["right"].set_color("#555555")
        ax.spines["right"].set_linewidth(0.9)
        ax.spines["top"].set_visible(False)
        ax.spines["bottom"].set_visible(False)
        ax.spines["left"].set_visible(False)
        ax.set_ylim(-AXIS_PAD, 1 + AXIS_PAD)
        ticks, texts = tick_spec(key, df[key].to_numpy(float))
        tick_positions = (np.asarray(ticks) - lows[j]) / spans[j]
        ax.set_yticks(tick_positions, texts, fontsize=8.5, color="#123b5d")
        ax.tick_params(axis="y", length=3.5, width=0.8, pad=3, colors="#123b5d")
        for tick_label in ax.get_yticklabels():
            tick_label.set_zorder(10)
            tick_label.set_bbox(dict(facecolor="white", edgecolor="none", alpha=0.88, pad=0.7))
        ax.set_ylabel(label, fontsize=9, color="#111111", rotation=0, labelpad=10, va="bottom")
        ax.yaxis.set_label_coords(j / (len(DISPLAY) - 1), 1.045, transform=host.transAxes)
        ax.patch.set_visible(False)
    title = "Hyperparameter exploration: best-MARE architecture highlighted" if highlighted else f"Hyperparameter exploration ({len(df)} architectures)"
    host.set_title(title, fontsize=15, color="#111111", pad=54, weight="semibold")
    sm = mpl.cm.ScalarMappable(norm=metric_norm, cmap=cmap)
    cax = fig.add_axes([0.925, 0.20, 0.012, 0.56])
    cb = fig.colorbar(sm, cax=cax)
    cb.set_label("log10(Test MARE)", fontsize=10)
    cb.ax.tick_params(labelsize=8)
    fig.patch.set_facecolor("white")
    host.set_facecolor("white")
    fig.subplots_adjust(left=0.06, right=0.90, top=0.80, bottom=0.07)
    fig.savefig(output.with_suffix(".pdf"), facecolor="white", bbox_inches="tight")
    fig.savefig(output.with_suffix(".png"), dpi=300, facecolor="white", bbox_inches="tight")
    plt.close(fig)


def plotly_dimensions(df):
    dimensions = []
    for key, label in DISPLAY:
        vals = df[key].to_numpy(float)
        ticks, texts = tick_spec(key, vals)
        lo, hi = np.nanmin(vals), np.nanmax(vals)
        span = hi - lo if hi > lo else 1.0
        dimensions.append(dict(label=label.replace("\n", " "), values=vals, range=[lo - AXIS_PAD * span, hi + AXIS_PAD * span], tickvals=ticks, ticktext=texts))
    return dimensions


def draw_html(df, output, highlighted=False):
    dims = plotly_dimensions(df)
    best = int(df["mare"].idxmin())
    traces = []
    if highlighted:
        traces.append(go.Parcoords(dimensions=dims, line=dict(color="#c9c9c9"), labelfont=dict(color="#111111"), tickfont=dict(color="#222222")))
        one = df.iloc[[best]]
        traces.append(go.Parcoords(dimensions=plotly_dimensions(one), line=dict(color=[one.iloc[0]["mare_log"]], colorscale="Viridis", cmin=df["mare_log"].min(), cmax=df["mare_log"].max(), showscale=True, colorbar=dict(title="log10(Test MARE)")), labelfont=dict(color="#111111"), tickfont=dict(color="#222222")))
    else:
        traces.append(go.Parcoords(dimensions=dims, line=dict(color=df["mare_log"], colorscale="Viridis", cmin=df["mare_log"].min(), cmax=df["mare_log"].max(), showscale=True, colorbar=dict(title="log10(Test MARE)")), labelfont=dict(color="#111111"), tickfont=dict(color="#222222")))
    title = "Hyperparameter exploration: best-MARE architecture highlighted" if highlighted else f"Hyperparameter exploration ({len(df)} architectures)"
    fig = go.Figure(traces)
    fig.update_layout(title=dict(text=title, x=0.5), width=1850, height=820, margin=dict(t=120, l=60, r=140, b=40), paper_bgcolor="white", plot_bgcolor="white", font=dict(color="#111111", size=12))
    fig.write_html(output.with_suffix(".html"), include_plotlyjs=True, full_html=True)


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("metadata", type=Path)
    parser.add_argument("output_dir", type=Path)
    args = parser.parse_args()
    args.output_dir.mkdir(parents=True, exist_ok=True)
    df = load_data(args.metadata)
    for stem, highlighted in (("parallel_hyperparams_thesis_all", False), ("parallel_hyperparams_thesis_best", True)):
        output = args.output_dir / stem
        draw_static(df, output, highlighted)
        draw_html(df, output, highlighted)
    best = df.loc[df["mare"].idxmin()]
    print(f"rows={len(df)} best={best['architecture']} MARE={best['mare']:.12g}")


if __name__ == "__main__":
    main()