augini
is a versatile Python framework that leverages AI for comprehensive data manipulation. It uses large language models to augment, generate, and anonymize tabular data, creating realistic and privacy-preserving datasets.
- Enhance existing datasets with AI-generated features
- Add contextual information based on current data
- Infuse domain knowledge from LLMs
- Create entirely new, realistic datasets
- Maintain statistical properties of original data
- Generate diverse, coherent synthetic profiles
- Implement k-anonymity and l-diversity
- Generate synthetic identifiers
- Balance privacy and data utility
- Augment ML training datasets
- Generate privacy-safe data for sharing
- Automatic labeling using state-of-the-art AI models
- Create synthetic data for software testing
- Develop realistic scenarios for business planning
- Produce diverse datasets for research and education
You can install Augini using pip:
pip install augini
Here's a simple example of how to use Augini:
from augini import Augini
import pandas as pd
api_key = "OpenAI or OpenRouter"
# OpenAI
augini = Augini(api_key=api_key, model='gpt-4-turbo', use_openrouter=False)
# OpenRouter
augini = Augini(api_key=api_key, use_openrouter=True, model='meta-llama/llama-3-8b-instruct')
# Create a sample DataFrame
data = {
'Place of Birth': ['New York', 'London', 'Tokyo'],
'Age': [30, 25, 40],
'Gender': ['Male', 'Female', 'Male']
}
df = pd.DataFrame(data)
# Add synthetic features
result_df = augini.augment_columns(df, ['NAME', 'OCCUPATION', 'FAVORITE_DRINK'])
print(result_df)
Custom prompts allow you to generate specific features based on your needs:
custom_prompt = "Based on the person's name and age, suggest a quirky pet for them. Respond with a JSON object with the key 'QuirkyPet'."
result_df = augini.augment_single(df, 'QuirkyPet', custom_prompt=custom_prompt)
print(result_df)
Leverage the knowledge embedded in language models to enhance your datasets:
description_prompt = "Generate a detailed description for a person based on their age and city. Respond with a JSON object with the key 'Description'."
result_df = augini.augment_single(df, 'Description', custom_prompt=description_prompt)
print(result_df)
recommendation_prompt = "Suggest a book and a movie for a person based on their age and city. Respond with a JSON object with keys 'RecommendedBook' and 'RecommendedMovie'."
result_df = augini.augment_single(df, 'Recommendations', custom_prompt=recommendation_prompt)
print(result_df)
You can anonymize sensitive information in your dataset by generating synthetic data:
from augini import Augini
import pandas as pd
api_key = "OpenAI or OpenRouter"
# OpenAI
augini = Augini(api_key=api_key, debug=False, use_openrouter=False, model='gpt-4-turbo')
# OpenRouter
augini = Augini(api_key=api_key, use_openrouter=True, model='meta-llama/llama-3-8b-instruct')
# Create a sample DataFrame with sensitive information
data = {
'Name': ['Alice Johnson', 'Bob Smith', 'Charlie Davis'],
'Age': [28, 34, 45],
'City': ['New York', 'Los Angeles', 'Chicago'],
'Email': ['[email protected]', '[email protected]', '[email protected]'],
'Phone': ['123-456-7890', '987-654-3210', '555-555-5555']
}
df = pd.DataFrame(data)
# Define a general anonymization prompt
anonymize_prompt = (
"Given the information from the dataset, create an anonymized version that protects individual privacy while maintaining data utility. "
"Follow these guidelines:\n\n"
"1. K-Anonymity: Ensure that each combination of quasi-identifiers (e.g., age, city) appears at least k times in the dataset. "
"Use generalization or suppression techniques as needed.\n"
"2. L-Diversity: For sensitive attributes, ensure there are at least l well-represented values within each equivalence class.\n"
"3. Direct Identifiers: Replace the following with synthetic data:\n"
" - Names: Generate culturally appropriate fictional names\n"
" - Email addresses: Create plausible fictional email addresses\n"
" - Phone numbers: Generate realistic but non-functional phone numbers\n"
"4. Quasi-Identifiers: Apply generalization or suppression as needed:\n"
" - Age: Consider using age ranges instead of exact ages\n"
" - City: Use broader geographic regions if necessary\n"
"5. Sensitive Attributes: Maintain the statistical distribution of sensitive data while ensuring diversity.\n"
"6. Data Consistency: Ensure that the anonymized data remains internally consistent and plausible.\n"
"7. Non-Sensitive Data: Keep unchanged unless required for k-anonymity or l-diversity.\n\n"
"Respond with a JSON object containing the anonymized values for all fields. "
"Ensure the anonymized dataset maintains utility for analysis while protecting individual privacy."
)
# Use the augment_columns method to anonymize the data
result_df = augini.augment_columns(df, ['Name_A', 'Email_A', 'Age_A', 'City_A'], custom_prompt=anonymize_prompt)
# Display the resulting DataFrame
print(result_df)
Output:
Name Age City Email Phone Name_A Email_A Age_A City_A
0 Alice Johnson 28 New York [email protected] 123-456-7890 Sophia Rodriguez [email protected] 25-30 East Coast
1 Bob Smith 34 Los Angeles [email protected] 987-654-3210 Sarah Johnson [email protected] 30-39 West Coast
2 Charlie Davis 45 Chicago [email protected] 555-555-5555 Emily Johnson [email protected] 40-50 Midwest
Augini can be used to automatically generate labels for data, enhancing datasets with semantic information. In this example, we use Augini to analyze sentences and generate semantic labels, sentiment analysis, and topic identification:
from augini import Augini
import pandas as pd
# Initialize Augini
api_key = "your_api_key_here"
augini = Augini(api_key=api_key, use_openrouter=True, model='gpt-4-turbo')
# Create a sample DataFrame with sentences
data = {
'sentence': [
"The cat sat on the mat.",
"I love to eat pizza on Fridays.",
"The stock market crashed yesterday.",
"She sang beautifully at the concert.",
"The new policy will be implemented next month."
]
}
df = pd.DataFrame(data)
# Define custom prompts for labeling
semantic_label_prompt = """
Analyze the given sentence and provide a semantic label. Choose from the following options:
Statement
Opinion
Fact
Action
Event
Respond with a JSON object containing the key 'semantic_label' and its value.
"""
sentiment_prompt = """
Determine the sentiment of the given sentence. Choose from the following options:
Positive
Negative
Neutral
Respond with a JSON object containing the key 'sentiment' and its value.
"""
topic_prompt = """
Identify the main topic of the given sentence. Provide a short (1-3 words) topic label.
Respond with a JSON object containing the key 'topic' and its value.
"""
# Generate labels using Augini
result_df = augini.augment_columns(df,
['semantic_label', 'sentiment', 'topic'],
custom_prompt=f"Sentence: {{sentence}}\n\n{semantic_label_prompt}\n\n{sentiment_prompt}\n\n{topic_prompt}"
)
# Display the results
print(result_df)
# You can also save the results to a CSV file
result_df.to_csv('labeled_sentences.csv', index=False)
Output:
| | sentence | semantic_label | sentiment | topic |
|---:|:-----------------------------------------------|:-----------------|:------------|:--------|
| 0 | The cat sat on the mat. | Statement | Neutral | Animal |
| 1 | I love to eat pizza on Fridays. | Opinion | Positive | Food |
| 2 | The stock market crashed yesterday. | Event | Negative | Finance |
| 3 | She sang beautifully at the concert. | Statement | Positive | Music |
| 4 | The new policy will be implemented next month. | Statement | Neutral | Policy |
If you are looking for an enterprise version of the tool or a need local version please contact us: