如何用Python/Pandas实现列转行及处理机器学习数据集标签映射

Hey there! Let's break down your two data processing questions with Pandas and get you sorted out.

If you're looking to split text within a single column into separate rows (like when each cell has multiple values separated by a delimiter such as commas or spaces), the explode() method paired with str.split() is your go-to tool. Here's a concrete example:

Example Code #

import pandas as pd

# Sample DataFrame with a single text column
df = pd.DataFrame({
    'text': ['apple, banana, cherry', 'dog; cat; rabbit', 'sun moon stars']
})

# Split the text by delimiter (adjust to match your data's separator) and explode into rows
df['text'] = df['text'].str.split(', ')  # Swap with '; ' or ' ' depending on your data
result_df = df.explode('text').reset_index(drop=True)

print(result_df)

Output #

text
0      apple
1     banana
2     cherry
3        dog
4        cat
5     rabbit
6        sun
7       moon
8      stars

If your column already has individual text entries that just need to be converted into a single column of rows (no splitting required), explode() still works—just wrap each value in a list first:

df['text'] = df['text'].apply(lambda x: [x])
result_df = df.explode('text')

Got it, let's tackle your machine learning dataset task. From your description:

• sol = 0 means sent0 violates common sense (label 0), so sent1 is valid (label 1)
• I’m assuming sol = 1 means sent1 violates common sense (label 0), so sent0 is valid (label 1)

Here are two straightforward methods to achieve this:

Method 1: Use melt() to Reshape and Add Labels #

This method converts your wide-form DataFrame to long-form first, then applies conditional logic to assign labels.

import pandas as pd

# Sample dataset matching your structure
df = pd.DataFrame({
    'sent0': ['The sun rises in the west', 'Water boils at 100°C', 'Plants need sunlight'],
    'sent1': ['The sun rises in the east', 'Water boils at 0°C', 'Plants need darkness'],
    'sol': [0, 1, 1]
})

# Reshape the DataFrame to long format
melted = df.melt(
    id_vars='sol',
    value_vars=['sent0', 'sent1'],
    var_name='sentence_source',
    value_name='sentence'
)

# Assign labels based on sol and sentence source
melted['label'] = melted.apply(
    lambda row: 0 if (row['sol'] == 0 and row['sentence_source'] == 'sent0') or (row['sol'] == 1 and row['sentence_source'] == 'sent1') else 1,
    axis=1
)

# Keep only the columns you need
final_df = melted[['sentence', 'label']].reset_index(drop=True)
print(final_df)

Method 2: Process Columns Separately and Concatenate #

If you prefer a more explicit approach, handle sent0 and sent1 individually then combine them.

import pandas as pd

# Same sample dataset
df = pd.DataFrame({
    'sent0': ['The sun rises in the west', 'Water boils at 100°C', 'Plants need sunlight'],
    'sent1': ['The sun rises in the east', 'Water boils at 0°C', 'Plants need darkness'],
    'sol': [0, 1, 1]
})

# Process sent0: label 0 if sol=0, else 1
sent0_df = df.rename(columns={'sent0': 'sentence'})
sent0_df['label'] = sent0_df['sol'].apply(lambda x: 0 if x == 0 else 1)

# Process sent1: label 1 if sol=0, else 0
sent1_df = df.rename(columns={'sent1': 'sentence'})
sent1_df['label'] = sent1_df['sol'].apply(lambda x: 1 if x == 0 else 0)

# Combine both DataFrames
final_df = pd.concat([sent0_df, sent1_df], ignore_index=True)[['sentence', 'label']]
print(final_df)

Output for Both Methods #

sentence  label
0     The sun rises in the west      0
1       Water boils at 100°C      1
2        Plants need sunlight      1
3     The sun rises in the east      1
4        Water boils at 0°C      0
5       Plants need darkness      0

Either method will get you the merged column with correct labels—pick whichever makes more sense for your workflow!

内容的提问来源于stack exchange，提问作者LeoCoder444