ai-data-science-team 0.0.0.9014

New Multi-Agent

• PandasDataAnalyst(): Combines Pandas Data Wrangling and Plotly Data Visualization Agents for performing data analysis and data visualization in a single agent.
• SQLDataAnalyst(): Includes a preprocessor that helps plan the steps and get better performance from SQLDatabaseAgent and DataVisualizationAgent.

Improvements

• Agents are all now imported at top level of ai_data_science_team. Users can now do from ai_data_science_team import SQLDatabaseAgent
• Agents now support langgraph checkpointer for state memory
• Agents now have name attribute

Full Changelog: 0.0.0.9013...0.0.0.9014

ai-data-science-team 0.0.0.9013

New App

• Exploratory Data Analysis Copilot: This application uses the EDAToolsAgent() to help the user create EDA Reports, Correlation Analysis, Missing Value Analysis, and general exploratory analysis. See App Here

Enhancements

• EDAToolsAgent(): New tool explain_data that returns a human-readable analysis of the data with statistical summary and various analysis that are common in exploratory data analysis.
• Tool Calling Agents now return state graphs with tool_calls captured as a list. This change helps the developer determine which tool was called last and how to handle the artifacts. This affectsEDAToolsAgent(), DataLoaderToolsAgent(), and MLflowToolsAgent().

Breaking Changes

• EDAToolsAgent():
  • The tool artifact dictionary keys have been updated. See Full Changelog below.
  • Sweetviz reports are not opened automatically unless the user specifies to do so.
  • Sweetviz reports are saved in a temporary directory.

Full Changelog: 0.0.0.9012...0.0.0.9013

ai-data-science-team 0.0.0.9012

New Agent

• EDA Tools Agent: Performs automated exploratory data analysis (EDA) with EDA Reporting, Missing Data Analysis, Correlation Analysis, and more. See Example

Full Changelog: 0.0.0.9011...0.0.0.9012

ai-data-science-team 0.0.0.9011

New Agents

• Data Loader Tools Agent: Loads data from various sources including CSV, Excel, Parquet, and Pickle files. See Example

Full Changelog: 0.0.0.9010...0.0.0.9011

ai-data-science-team 0.0.0.9010

New Agents

• MLflowToolsAgent: This agent has 11+ tools for managing models, ML projects, and making production ML predictions with MLflow.
• New Example: MLflow Agent See Example

New AI Apps

• Created app/ directory: Houses AI applications that demonstrate usage of the AI Data Science Team
• SQL Database Agent App: Connects any SQL Database, generates SQL queries from natural language, and returns data as a downloadable table. See Application

Internal Changes

• Refactored utils, parsers, and tools to make it more clear the function roles
• Async updates

Full Changelog: 0.0.0.9009...0.0.0.9010

ai-data-science-team 0.0.0.9009

New Agents:

• H2OMLAgent(): The first in a series of ML agents designed to make Machine Learning Models with AI. This AI Agent is trained in h2o AutoML and is capable of creating 100's of ML models in seconds.
• New Example: https://github.com/business-science/ai-data-science-team/blob/master/examples/ml_agents/h2o_machine_learning_agent.ipynb

Improvements

• Workflow Summary Report: The explain code step was replaced with a much faster step for documenting the agentic workflow. A get_workflow_summary() method returns formatted summary reports of every step taken in the agentic workflow.
• Smart Schema Pruning: SQL Database Agent gained a new parameter, smart_schema_pruning, which uses an extra LLM call to prune tables and columns. This is useful when database schemas are very large. Pruning is based on Uber QueryGPT blog article which implements a Column Prune Agent. Read more here: https://www.uber.com/blog/query-gpt/

Full Changelog: 0.0.0.9008...0.0.0.9009

ai-data-science-team 0.0.0.9008

New Features

1. New Object-Oriented Programming Framework (Experimental): OOP Framework provides a Pythonic interface to agents, improved methods, and more features beyond LangGraph methods. New classes include DataCleaningAgent(), FeatureEngineeringAgent(), SQLDatabaseAgent() and more.
2. Multi-Agents: A new multiagents module was created. This supports common LangGraph multi-agent architectures, which will be a big focus going forward.
3. New SQLDataAnalyst Multi-Agent: Combines the SQLDatabaseAgent and DataVisualizationAgent() in a multi-agent workflow with conditional routing to the data visualization agent. Perfect for Business Intelligence and Data Analysis applications.

New Examples

1. How to Build SQL Data Analysis Agents: https://github.com/business-science/ai-data-science-team/blob/master/examples/multiagents/sql_data_analyst.ipynb
2. Human In The Loop (new workflow): https://github.com/business-science/ai-data-science-team/blob/master/examples/advanced_topics/human_in_the_loop.ipynb

Enhancements

• New BaseAgent() Class: Used to make common methods available to all OOP agents.
• New Human-In-The-Loop Workflow: Allows applications to include human review and modification. Perfect for iteratively improving AI functions.

Full Changelog: 0.0.0.9007...0.0.0.9008

ai-data-science-team 0.0.0.9007

New Agent:

• Data Visualization Agent: Generates code for data visualizations
• New Example: Automate Data Visualization with AI Agents https://github.com/business-science/ai-data-science-team/blob/master/examples/data_visualization_agent.ipynb

Agent Enhancements:

• Add n_samples to allow users to control the number of data rows passed to LLM prompts.
• Add file_name to allow users to control the file name that the agent uses when logging functions.
• plotly_from_dict() helper utility to convert a dictionary to a plotly graph.

Fixes:

• Bypass steps - Adds all_datasets_summary_str to allow LLM to know the dataset summary if the recommendation step is bypassed.
• Improved get_database_metadata() for SQL engine.

Full Changelog: 0.0.0.9005...0.0.0.9007

ai-data-science-team 0.0.0.9005

New Agents

• SQL Database Agent: Queries SQL databases from Natural Language, automates pipelines as import-ready python functions, and makes it easy to integrate SQL agents into Streamlit apps
• New Tutorial: https://github.com/business-science/ai-data-science-team/blob/master/examples/sql_database_agent.ipynb

Enhancements

• Dynamically Bypass Long-Running Steps: This is important if speed is critical. Planning steps (e.g. recommend steps for coding agent, explaining code step) can be bypassed to reduce LLM calls and speed up operations.

Full Changelog: 0.0.0.9004...0.0.0.9005

ai-data-science-team 0.0.0.9004

New Agent:

• Data Wrangling Agent - Handles multiple datasets, merges, joins, and prepares data for analysis
• New Example: How to automate data wrangling with AI

Full Changelog: 0.0.0.9003...0.0.0.9004