Set log axis for some parameters

[ddiama]

Hello,

I have the code below to create a parallel coordinates plot

paxfig = paxplot.pax_parallel(n_axes=len(cols))
paxfig.plot(df.to_numpy())

paxfig.set_labels(column_names)

paxfig.set_size_inches(12, 7)
paxfig.subplots_adjust(left=0.1, bottom=0.2, right=0.9, top=0.9)

color_col = 5
paxfig.add_colorbar(
ax_idx=color_col,
cmap='jet',
colorbar_kwargs={'label': r"$L_{MSE}$ Validation"}
)

Is it possible to have for some of the parameters (e.g. the last two) log scale in the axis in order to be able to see them better?

[kravitsjacob]

Hello @ddiama!

Great question. There is currently no native support for log scales in paxplot. However, you can use the custom tick label functionality with data scaling as a workaround:

import matplotlib.pyplot as plt
import pandas as pd
import paxplot
import math

# Original data (trying to mimic your data's behavior)
data = [
    [1.0, 1.0],
    [1.1, 1.1],
    [5.0, 1.3],
    [9.0, 1.5],
    [10.0, 1.6],
    [50.0, 1.7],
    [100.0, 1.8],
]
df = pd.DataFrame(data, columns=['A', 'B'])

# Log transform data
log_scale_func = math.log10
df['Log Version A'] = df['A'].apply(log_scale_func)

# Create figure
cols = df.columns
paxfig = paxplot.pax_parallel(n_axes=len(cols))
paxfig.plot(df.to_numpy())
paxfig.set_labels(cols)
paxfig.set_ticks(
    ax_idx=2,
    ticks=[log_scale_func(1), log_scale_func(10), log_scale_func(100)],
    labels=['1', '10', '100']
)
plt.show()

Adding native log support would not be too hard, and I would be happy to support any interest in adding it!

Also, very shameless plug here based on your labels/data. you may be interested in this blog post I wrote on multi-objective hyperparamter tuning. This blog post also links to other publications.