Curve Plots

Curve plots show how a metric evolves over a continuous x-axis field (typically epoch or step). This is the most common plot type for visualizing training dynamics.

Basic usage

malet-plot -exp_folder ./experiments/my_exp \
  -mode curve-epoch-val_accuracy \
  -best_at_max

This plots val_accuracy on the y-axis over epoch on the x-axis. Seeds are averaged automatically, and all other hyperparameters are optimized (best value selected).

Error bands and error bars

When averaging over seeds, Malet computes standard error and displays it as shaded bands by default. You can switch between display modes:

• std_plot: fill — shaded band (default)

• std_plot: bar — error bars

• std_plot: none — no error display

Curve with best point starred

Use curve_best mode to highlight the best-performing point with a star marker:

malet-plot -exp_folder ./experiments/my_exp \
  -mode curve_best-epoch-val_accuracy \
  -best_at_max

This is useful for identifying the epoch where a model peaks before overfitting begins.

You can also use curve_best with a hyperparameter on the x-axis to find the optimal hyperparameter value:

malet-plot -exp_folder ./experiments/my_exp \
  -mode curve_best-rho-val_accuracy \
  -filter 'step last' \
  -best_at_max

Python API

For fine-grained control, use ax_draw_curve and ax_draw_best_stared_curve directly:

import matplotlib.pyplot as plt
from malet.experiment import ExperimentLog
from malet.plot_utils.data_processor import avgbest_df, select_df
from malet.plot_utils.plot_drawer import ax_draw_curve

log = ExperimentLog.from_tsv('log.tsv')
df = log.melt_and_explode_metric()
df = select_df(df, {'metric': 'val_accuracy'})
df = avgbest_df(df, 'metric_value', avg_over={'seed'},
                best_over={'lr'}, best_at_max=True)

# Reduce to single index (step)
drop = [n for n in df.index.names if n != 'step']
df = df.reset_index(drop, drop=True).sort_index()

fig, ax = plt.subplots(figsize=(9, 6))
ax_draw_curve(ax, df, label='ResNet20', color='#1f77b4',
              annotate=False, std_plot='fill')
ax.set_xlabel('Epoch')
ax.set_ylabel('Val Accuracy')
ax.legend()
fig.savefig('curve.png', dpi=150, bbox_inches='tight')

ax_draw_curve parameters

Parameter	Type	Default	Description
ax	Axes	—	Matplotlib axes to draw on
df	DataFrame	—	Single-index DataFrame with one value column
label	str	—	Legend label
std_plot	str	'fill'	Error display: 'fill', 'bar', or 'none'
color	str	'orange'	Line color
linewidth	float	4	Line width
marker	str	'D'	Marker style
markersize	float	10	Marker size
markevery	int	20	Show marker every N points
linestyle	str	'-'	Line style
annotate	bool	True	Add value annotations
unif_xticks	bool	False	Uniform x-tick spacing

When the number of data points exceeds 100, markers are disabled automatically.