Training Loss Curve — Pygal

A line plot showing training and validation loss curves over epochs during neural network training. This visualization is essential for monitoring model training, detecting overfitting (when validation loss diverges from training loss), and determining optimal early stopping points. The dual-curve display reveals the gap between training and generalization performance.

Training Loss Curve rendered with Pygal

Python source (Pygal)

""" anyplot.ai
line-loss-training: Training Loss Curve
Library: pygal 3.1.0 | Python 3.13.13
Quality: 92/100 | Updated: 2026-05-14
"""

import os
import sys

import numpy as np


# Remove local directory from sys.path to avoid shadowing the pygal package
_local_dir = os.path.abspath(os.path.dirname(__file__))
sys.path = [p for p in sys.path if os.path.abspath(p) != _local_dir]

import pygal
from pygal.style import Style


THEME = os.getenv("ANYPLOT_THEME", "light")

# Theme tokens (from default-style-guide.md)
PAGE_BG = "#FAF8F1" if THEME == "light" else "#1A1A17"
INK = "#1A1A17" if THEME == "light" else "#F0EFE8"
INK_MUTED = "#6B6A63" if THEME == "light" else "#A8A79F"

# Okabe-Ito palette (first series = brand green #009E73)
IMPRINT = ("#009E73", "#C475FD", "#4467A3", "#BD8233", "#AE3030", "#2ABCCD", "#954477")

# Data: Simulated training loss curves showing typical overfitting behavior
np.random.seed(42)
epochs = np.arange(1, 51)

# Training loss: Steadily decreasing with some noise
train_loss = 2.5 * np.exp(-0.08 * epochs) + 0.1 + np.random.normal(0, 0.02, len(epochs))

# Validation loss: Decreases then increases (overfitting after epoch ~25)
val_loss = 2.3 * np.exp(-0.07 * epochs) + 0.15 + 0.003 * np.maximum(0, epochs - 25) ** 1.5
val_loss += np.random.normal(0, 0.03, len(epochs))

# Find minimum validation loss epoch for annotation
min_val_epoch = int(epochs[np.argmin(val_loss)])
min_val_loss = float(np.min(val_loss))

# Custom theme-adaptive style
custom_style = Style(
    background=PAGE_BG,
    plot_background=PAGE_BG,
    foreground=INK,
    foreground_strong=INK,
    foreground_subtle=INK_MUTED,
    colors=IMPRINT,
    title_font_size=28,
    label_font_size=22,
    major_label_font_size=18,
    legend_font_size=16,
    value_font_size=14,
    stroke_width=3,
)

# Create line chart
chart = pygal.Line(
    width=4800,
    height=2700,
    style=custom_style,
    title="line-loss-training · pygal · anyplot.ai",
    x_title="Epoch",
    y_title="Cross-Entropy Loss",
    show_x_guides=True,
    show_y_guides=True,
    dots_size=6,
    stroke_style={"width": 3},
    legend_at_bottom=False,
    legend_box_size=24,
    margin=80,
    x_label_rotation=0,
    truncate_label=-1,
    show_dots=True,
)

# Set x-axis labels (show every 5th epoch for readability)
chart.x_labels = [str(e) if e % 5 == 0 else "" for e in epochs]

# Add training and validation loss data
chart.add("Training Loss", list(train_loss))
chart.add("Validation Loss", list(val_loss))

# Save as PNG and HTML with theme suffix
chart.render_to_png(f"plot-{THEME}.png")
with open(f"plot-{THEME}.html", "wb") as f:
    f.write(chart.render())

Part of Training Loss Curve on anyplot.ai.

Other implementations