是否有更快/更少的 RAM 使用方式来使用 Python 汇集数据？

Question

enter image description here

https://kin-phinf.pstatic.net/20221001_267/1664597566757fY2pz_PNG/%C8%AD%B8%E9_%C4%B8%C3%B3_2022-10-01_001049.png?type=w750

I want to pool a data like the figure above, but it takes too much time and RAM usage. Can I make it faster / efficient?

My code is like this:

data = df.groupby(['Name', 'Age', 'Pet', 'Allergy']).apply(lambda x: pd.Series(range(x['Amount'].squeeze()))).reset_index()
data = df.groupby(['Name', 'Age', 'Pet', 'Allergy']).apply(lambda x: pd.Series(range(x['Amount'].squeeze()))).reset_index()[['Name', 'Age', 'Pet', 'Allergy']]

enter image description here It's kind of an abbreviated form, but my actual dataset is 3.5GB.. So it takes really long time. I wonder if there's any other way to do this work more fast.

I'd appreciate any help! Thank you!

Answer 1

You could preallocate the final dataframe, then iterate the original dataframe, reassigning rows in the final.

import pandas as pd
import numpy as np

df = pd.DataFrame({"Name":["Male", "Female"],
    "Age":[29, 43], "Pet":["Cat", "Dog"],
    "Allergy":["Negative", "Positive"],
    "Amount":[2, 4]})

amounts = df["Amount"]
df.drop("Amount", axis=1, inplace=True)
counts = amounts.sum()

new_df = pd.DataFrame(columns=df.columns, index=np.arange(counts))
new_index = 0

for amount, (_, row) in zip(amounts, df.iterrows()):
    for i in range(new_index, new_index+amount):
        new_df.iloc[i] = row
    new_index = new_index+amount

del df, amounts, row

print(new_df)