Intelligent Fill with the ai.fillna Method¶

This notebook explains the feature in order: the main input, the output, and the practical benefits.

1. Setup¶

Configure authentication and choose a responses model.

In [1]:

import os

import pandas as pd

from openaivec import pandas_ext

assert os.getenv("OPENAI_API_KEY") or os.getenv("AZURE_OPENAI_BASE_URL"), (
    "Set OPENAI_API_KEY or Azure OpenAI environment variables before running this notebook."
)

pandas_ext.set_responses_model("gpt-4.1-mini")

import os import pandas as pd from openaivec import pandas_ext assert os.getenv("OPENAI_API_KEY") or os.getenv("AZURE_OPENAI_BASE_URL"), ( "Set OPENAI_API_KEY or Azure OpenAI environment variables before running this notebook." ) pandas_ext.set_responses_model("gpt-4.1-mini")

2. Input: DataFrame and target column¶

The input is a normal DataFrame with missing values in one target column.

In [2]:

products = pd.DataFrame({
    "product": ["Wireless Earbuds", "Yoga Mat", "Desk Lamp", "Water Bottle", "Running Shoes"],
    "category": ["Electronics", "Fitness", "Home", "Fitness", "Footwear"],
    "price_usd": [129.0, 35.0, 48.0, 20.0, 95.0],
    "marketing_copy": [
        "Immersive sound with all-day battery life",
        None,
        "Warm light for focused evening work",
        None,
        None,
    ],
})

products

products = pd.DataFrame({ "product": ["Wireless Earbuds", "Yoga Mat", "Desk Lamp", "Water Bottle", "Running Shoes"], "category": ["Electronics", "Fitness", "Home", "Fitness", "Footwear"], "price_usd": [129.0, 35.0, 48.0, 20.0, 95.0], "marketing_copy": [ "Immersive sound with all-day battery life", None, "Warm light for focused evening work", None, None, ], }) products

Out[2]:

	product	category	price_usd	marketing_copy
0	Wireless Earbuds	Electronics	129.0	Immersive sound with all-day battery life
1	Yoga Mat	Fitness	35.0	None
2	Desk Lamp	Home	48.0	Warm light for focused evening work
3	Water Bottle	Fitness	20.0	None
4	Running Shoes	Footwear	95.0	None

3. Run intelligent fill¶

Call .ai.fillna() with the target column. The method fills only missing values in that column.

In [3]:

filled_products = products.ai.fillna(
    target_column_name="marketing_copy",
    batch_size=16,
    show_progress=True,
)

filled_products

filled_products = products.ai.fillna( target_column_name="marketing_copy", batch_size=16, show_progress=True, ) filled_products

Processing batches:   0%|          | 0/3 [00:00<?, ?item/s]

Out[3]:

	product	category	price_usd	marketing_copy
0	Wireless Earbuds	Electronics	129.0	Immersive sound with all-day battery life
1	Yoga Mat	Fitness	35.0	Non-slip surface for safe and comfortable work...
2	Desk Lamp	Home	48.0	Warm light for focused evening work
3	Water Bottle	Fitness	20.0	Durable and lightweight bottle to stay hydrate...
4	Running Shoes	Footwear	95.0	Lightweight and supportive design for optimal ...

4. Output: before and after¶

The output keeps the same schema and index, with missing target values completed.

In [4]:

summary = products[["product", "marketing_copy"]].rename(columns={"marketing_copy": "before"}).assign(
    after=filled_products["marketing_copy"]
)
summary

summary = products[["product", "marketing_copy"]].rename(columns={"marketing_copy": "before"}).assign( after=filled_products["marketing_copy"] ) summary

Out[4]:

	product	before	after
0	Wireless Earbuds	Immersive sound with all-day battery life	Immersive sound with all-day battery life
1	Yoga Mat	None	Non-slip surface for safe and comfortable work...
2	Desk Lamp	Warm light for focused evening work	Warm light for focused evening work
3	Water Bottle	None	Durable and lightweight bottle to stay hydrate...
4	Running Shoes	None	Lightweight and supportive design for optimal ...

5. Benefits¶

Main input

• A DataFrame
• One target column name that contains missing values

Main output

• A DataFrame with the same rows and columns
• Missing values in the target column filled with context-aware values

Why this is useful

• Uses row context from other columns instead of simple mean/mode rules
• Keeps the pandas workflow simple (df.ai.fillna(...))
• Supports batching controls (batch_size) for larger datasets