plotting.py

A lightweight Plotly wrapper for fast prototyping of common scientific and ML charts — minimal boilerplate, consistent styling out of the box.

Installation

pip install plotly kaleido scikit-learn matplotlib numpy

Usage

from plotting import plotBox, plotCompare, plotHeatmap  # etc.

Every function accepts a title, x_label / y_label, an optional size=(height, width) tuple, and two output modes:

Functions

plotBox — grouped box plots

plotBox(
    {'Model A': array_a, 'Model B': array_b},
    title='Accuracy by Model',
    y_label='Accuracy',
    ylimit=[0.5, 1.0],
)

plotCompare — line chart with shared x axis

Best for training curves, time series, anything where all series share the same x values.

epochs = np.arange(1, 51)
plotCompare(
    epochs,
    {'Train': train_loss, 'Val': val_loss},
    title='Loss Curves',
    x_label='Epoch', y_label='Loss',
)

plotCombine — multi-series with independent x arrays

Use when each series has its own x domain or resolution.

plotCombine(
    {'sin': (t, np.sin(t)), 'cos': (t, np.cos(t))},
    title='Trig Functions',
    mode='lines',  # 'markers' or 'lines+markers' also accepted
)

plotScatter — scatter plot, shared x

plotScatter(
    x_vals,
    {'Group A': y_a, 'Group B': y_b},
    title='Feature vs Target',
    dot_size=4,
)

plotMap — raw heatmap (no interpolation)

Good for confusion matrices, correlation grids, any discrete 2-D matrix.

plotMap(z=matrix, x=col_labels, y=row_labels, title='Confusion Matrix')

plotHeatmap — contour plot with heatmap fill

For continuous surfaces where you want to highlight level sets.

plotHeatmap(z=Z, x=grid_x, y=grid_y, title='Loss Surface', ncontours=15)

plotTrajectory — 2-D paths with time-varying marker size

Markers grow over time so you can track direction at a glance.

plotTrajectory(
    evo=np.arange(T),           # drives marker size (normalised to [0, 15])
    plot={'Agent': path},       # path is an (N, 2) array
    title='Agent Trajectory',
)

plotOptimization — loss landscape + optimiser paths

Overlays one or more optimiser trajectories on a filled contour map.

plotOptimization(
    z=loss_surface, x=w1, y=w2,
    paths={'SGD': sgd_path, 'Adam': adam_path},
    title='Optimiser Comparison',
)

viewPCA — 2-D PCA projection (matplotlib)

Reduces high-dimensional data to 2 components and colours points by a given index.

viewPCA(X, index=labels, title='PCA Projection', x='PC 1', y='PC 2')

Exporting

# Save a PNG (written twice to avoid kaleido race condition)
plotCompare(epochs, data, title='loss', export='png')

# Save an interactive HTML file
plotCompare(epochs, data, title='loss', html=True)