Plotly Ignores Colors

• Weights and Biases version: ‘0.8.21’
• Python version: 3.5.3
• Operating System: GCP AI Notebook with Tensorflow 2.0 Environment Jupyter Lab

Description

If I create a bar plot using matplotlib.pyplot and pass in a set of colors to use for each bar, it displays the plot correctly with the corresponding colors. When I log this plot to wandb, the plot displayed in wandb shows the correct bars but all the same color. Alternatively, if I log the plot to wandb as an image, it uploads the plot correctly with the corresponding colors.

What I Did

import numpy as np
import matplotlib.pylplot as plt
import seaborn as sns

lengths, modes = np.unique(counts, return_counts=True)
pal = list(sns.color_palette("hls", len(lengths)))
plt.bar(lengths[1:], modes[1:], color=pal)
plt.xlabel('Visit Lengths')
plt.ylabel('Frequency')
wandb.log({'Visit Length Distribution' : plt})

[vanpelt]

I have a branch that fixes this, you can use with: pip install --upgrade git+git://github.com/wandb/client.git@bug/plotly_np#egg=wandb we’ll get this out in the next release.

[lorinczszabolcs]

Hi! I was facing a similar issue, when I created a seaborn scatter plot. When logging as a wandb.Image, it showed the colors, but when I logged the figure directly, the colors were not shown on the legend, only on the actual points in the scatter plot.

Example:

from matplotlib import pyplot as plt
import seaborn as sns

def _get_color_mapping():
    return {
        'tree': '#124A54', 'sky': '#28A2B8', 'bear': '#58C9DA',
        'bee': '#ADE68E', 'honey': '#CBFCB1'}

classes = np.array(['tree', 'sky', 'bear', 'bee', 'honey'])
fig, _ = plt.subplots(figsize=(10., 10.))
sns.scatterplot(x=np.random.random(1000), y=np.random.random(1000), hue=classes[np.random.randint(0, 5, (1000))], palette=_get_color_mapping())

wandb.log({
        "Figures/scatter": wandb.Image(fig) # works
        # "Figures/scatter": fig # colors do not appear in legend
        })

For info:

plotly==5.3.1 matplotlib==3.3.4 seaborn==0.11.2