如何使用Pandas按model与scheduler两列分组绘制mae列的多组条形图？

Hey there! The issue with your current code is that groupby().plot.bar() creates a separate subplot for each (model, scheduler) pair instead of grouping bars by model as you want. Let's fix this with two straightforward approaches:

Approach 1: Use Seaborn (Simplest for Grouped Plots) #

Seaborn's catplot is built exactly for this kind of categorical grouped visualization. Here's how to use it:

First, let's make sure we have your DataFrame set up correctly (I'll recreate it from your table):

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

# Recreate your DataFrame
data = [
    ["ecaresnet50t", "warm", 4.518],
    ["ecaresnet50t", "cosine", 4.46],
    ["ecaresnet50t", "constant", 4.972],
    ["resnest50d", "warm", 4.056],
    ["resnest50d", "cosine", 4.1],
    ["resnest50d", "constant", 5.072],
    ["resnetrs50", "warm", 4.164],
    ["resnetrs50", "cosine", 4.154],
    ["resnetrs50", "constant", 4.644],
    ["seresnet50", "warm", 4.202]
]
df = pd.DataFrame(data, columns=["model", "scheduler", "mae"])

Now plot the grouped bar chart:

# Create the grouped bar plot
sns.catplot(
    x="model",
    y="mae",
    hue="scheduler",
    kind="bar",
    data=df,
    palette="viridis"
)

# Customize the plot for readability
plt.title("MAE Scores by Model and Scheduler")
plt.xlabel("Model")
plt.ylabel("MAE Score")
plt.xticks(rotation=45)  # Rotate model names to avoid overlap
plt.tight_layout()  # Prevent label cutoff
plt.show()

This will generate a single plot where each model has 3 distinct bars (one for each scheduler), colored differently for easy comparison.

Approach 2: Use Pandas Pivot + Plot #

If you prefer sticking with pandas, reshape your DataFrame into a wide format first using pivot_table, then plot the bars:

# Reshape the DataFrame to wide format (model as rows, scheduler as columns)
pivot_df = df.pivot_table(index="model", columns="scheduler", values="mae")

# Plot the grouped bars
pivot_df.plot(kind="bar", figsize=(10,6), palette="viridis")

# Customize the plot
plt.title("MAE Scores by Model and Scheduler")
plt.xlabel("Model")
plt.ylabel("MAE Score")
plt.xticks(rotation=45)
plt.legend(title="Scheduler")
plt.tight_layout()
plt.show()

This method organizes your data so each model is a row, each scheduler is a column, then plots each column as a bar group for the corresponding model.

Both approaches will give you the grouped bar chart you're looking for—pick whichever fits your workflow better!

内容的提问来源于stack exchange，提问作者Talha Anwar