ai-data-science-team 0.0.0.9000__py3-none-any.whl → 0.0.0.9005__py3-none-any.whl
Sign up to get free protection for your applications and to get access to all the features.
- ai_data_science_team/_version.py +1 -1
- ai_data_science_team/agents/__init__.py +4 -0
- ai_data_science_team/agents/data_cleaning_agent.py +347 -0
- ai_data_science_team/agents/data_wrangling_agent.py +365 -0
- ai_data_science_team/agents/feature_engineering_agent.py +368 -0
- ai_data_science_team/templates/__init__.py +0 -0
- ai_data_science_team/templates/agent_templates.py +409 -0
- ai_data_science_team/tools/__init__.py +0 -0
- ai_data_science_team/tools/data_analysis.py +116 -0
- ai_data_science_team/tools/logging.py +61 -0
- ai_data_science_team/tools/parsers.py +57 -0
- ai_data_science_team/tools/regex.py +73 -0
- ai_data_science_team-0.0.0.9005.dist-info/METADATA +162 -0
- ai_data_science_team-0.0.0.9005.dist-info/RECORD +19 -0
- ai_data_science_team/agents.py +0 -325
- ai_data_science_team-0.0.0.9000.dist-info/METADATA +0 -131
- ai_data_science_team-0.0.0.9000.dist-info/RECORD +0 -9
- {ai_data_science_team-0.0.0.9000.dist-info → ai_data_science_team-0.0.0.9005.dist-info}/LICENSE +0 -0
- {ai_data_science_team-0.0.0.9000.dist-info → ai_data_science_team-0.0.0.9005.dist-info}/WHEEL +0 -0
- {ai_data_science_team-0.0.0.9000.dist-info → ai_data_science_team-0.0.0.9005.dist-info}/top_level.txt +0 -0
@@ -0,0 +1,73 @@
|
|
1
|
+
import re
|
2
|
+
from datetime import datetime
|
3
|
+
|
4
|
+
|
5
|
+
def relocate_imports_inside_function(code_text):
|
6
|
+
"""
|
7
|
+
Relocates all import statements in a given Python function and moves them inside the function definition.
|
8
|
+
|
9
|
+
Parameters
|
10
|
+
----------
|
11
|
+
code_text : str
|
12
|
+
The Python code as a string.
|
13
|
+
|
14
|
+
Returns
|
15
|
+
-------
|
16
|
+
str
|
17
|
+
The modified Python code with imports relocated inside the function.
|
18
|
+
"""
|
19
|
+
# Match all import statements
|
20
|
+
import_pattern = r'^\s*(import\s+[^\n]+|from\s+\S+\s+import\s+[^\n]+)\s*$'
|
21
|
+
imports = re.findall(import_pattern, code_text, re.MULTILINE)
|
22
|
+
|
23
|
+
# Remove imports from the top-level code
|
24
|
+
code_without_imports = re.sub(import_pattern, '', code_text, flags=re.MULTILINE).strip()
|
25
|
+
|
26
|
+
# Find the function definition and insert the imports inside it
|
27
|
+
function_pattern = r'(def\s+\w+\s*\(.*?\):)'
|
28
|
+
match = re.search(function_pattern, code_without_imports)
|
29
|
+
|
30
|
+
if match:
|
31
|
+
function_start = match.end()
|
32
|
+
# Insert the imports right after the function definition
|
33
|
+
imports_code = '\n ' + '\n '.join(imports) # Indent imports
|
34
|
+
modified_code = (
|
35
|
+
code_without_imports[:function_start]
|
36
|
+
+ imports_code
|
37
|
+
+ code_without_imports[function_start:]
|
38
|
+
)
|
39
|
+
return modified_code
|
40
|
+
|
41
|
+
# If no function is found, return the original code
|
42
|
+
return code_text
|
43
|
+
|
44
|
+
def add_comments_to_top(code_text, agent_name="data_wrangler"):
|
45
|
+
"""
|
46
|
+
Adds AI-generated metadata comments to the top of the Python code.
|
47
|
+
|
48
|
+
Parameters
|
49
|
+
----------
|
50
|
+
code_text : str
|
51
|
+
The Python code to be commented.
|
52
|
+
agent_name : str, optional
|
53
|
+
The agent name to include in the comments, by default "data_wrangler".
|
54
|
+
|
55
|
+
Returns
|
56
|
+
-------
|
57
|
+
str
|
58
|
+
The Python code with the added comments at the top.
|
59
|
+
"""
|
60
|
+
# Generate timestamp
|
61
|
+
time_created = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
|
62
|
+
|
63
|
+
# Construct the header comments
|
64
|
+
header_comments = [
|
65
|
+
"# Disclaimer: This function was generated by AI. Please review before using.",
|
66
|
+
f"# Agent Name: {agent_name}",
|
67
|
+
f"# Time Created: {time_created}",
|
68
|
+
""
|
69
|
+
]
|
70
|
+
|
71
|
+
# Join the header with newlines, then prepend to the existing code_text
|
72
|
+
header_block = "\n".join(header_comments)
|
73
|
+
return header_block + code_text
|
@@ -0,0 +1,162 @@
|
|
1
|
+
Metadata-Version: 2.1
|
2
|
+
Name: ai-data-science-team
|
3
|
+
Version: 0.0.0.9005
|
4
|
+
Summary: Build and run an AI-powered data science team.
|
5
|
+
Home-page: https://github.com/business-science/ai-data-science-team
|
6
|
+
Author: Matt Dancho
|
7
|
+
Author-email: mdancho@business-science.io
|
8
|
+
Requires-Python: >=3.9
|
9
|
+
Description-Content-Type: text/markdown
|
10
|
+
License-File: LICENSE
|
11
|
+
Requires-Dist: openpyxl
|
12
|
+
Requires-Dist: langchain
|
13
|
+
Requires-Dist: langchain_community
|
14
|
+
Requires-Dist: langchain_openai
|
15
|
+
Requires-Dist: langchain_experimental
|
16
|
+
Requires-Dist: langgraph>=0.2.57
|
17
|
+
Requires-Dist: openai
|
18
|
+
Requires-Dist: pandas
|
19
|
+
Requires-Dist: numpy
|
20
|
+
Requires-Dist: plotly
|
21
|
+
Requires-Dist: streamlit
|
22
|
+
Requires-Dist: scikit-learn
|
23
|
+
Requires-Dist: xgboost
|
24
|
+
|
25
|
+
# Your AI Data Science Team (An Army Of Copilots)
|
26
|
+
|
27
|
+
**An AI-powered data science team of copilots that uses agents to help you perform common data science tasks 10X faster**.
|
28
|
+
|
29
|
+
Star ⭐ This GitHub (Takes 2 seconds and means a lot).
|
30
|
+
|
31
|
+
---
|
32
|
+
|
33
|
+
The AI Data Science Team of Copilots includes Agents that specialize data cleaning, preparation, feature engineering, modeling (machine learning), and interpretation of various business problems like:
|
34
|
+
|
35
|
+
- Churn Modeling
|
36
|
+
- Employee Attrition
|
37
|
+
- Lead Scoring
|
38
|
+
- Insurance Risk
|
39
|
+
- Credit Card Risk
|
40
|
+
- And more
|
41
|
+
|
42
|
+
## Companies That Want An AI Data Science Team Copilot
|
43
|
+
|
44
|
+
If you are interested in having your own custom enteprise-grade AI Data Science Team Copilot, send inquiries here: [https://www.business-science.io/contact.html](https://www.business-science.io/contact.html)
|
45
|
+
|
46
|
+
## Free Generative AI For Data Scientists Workshop
|
47
|
+
|
48
|
+
If you want to learn how to build AI Agents for your company that performs Data Science, Business Intelligence, Churn Modeling, Time Series Forecasting, and more, [register for my next Generative AI for Data Scientists workshop here.](https://learn.business-science.io/ai-register)
|
49
|
+
|
50
|
+
## Data Science Agents
|
51
|
+
|
52
|
+
This project is a work in progress. New data science agents will be released soon.
|
53
|
+
|
54
|
+

|
55
|
+
|
56
|
+
### Agents Available Now
|
57
|
+
|
58
|
+
1. **Data Wrangling Agent:** Merges, Joins, Preps and Wrangles data into a format that is ready for data analysis.
|
59
|
+
2. **Data Cleaning Agent:** Performs Data Preparation steps including handling missing values, outliers, and data type conversions.
|
60
|
+
3. **Feature Engineering Agent:** Converts the prepared data into ML-ready data. Adds features to increase predictive accuracy of ML models.
|
61
|
+
|
62
|
+
### Agents Coming Soon
|
63
|
+
|
64
|
+
1. **Data Analyst:** Analyzes data structure, creates exploratory visualizations, and performs correlation analysis to identify relationships.
|
65
|
+
2. **Machine Learning Agent:** Builds and logs the machine learning models.
|
66
|
+
3. **Interpretability Agent:** Performs Interpretable ML to explain why the model returned predictions including which features were the most important to the model.
|
67
|
+
4. **Supervisor:** Forms task list. Moderates sub-agents. Returns completed assignment.
|
68
|
+
|
69
|
+
## Disclaimer
|
70
|
+
|
71
|
+
**This project is for educational purposes only.**
|
72
|
+
|
73
|
+
- It is not intended to replace your company's data science team
|
74
|
+
- No warranties or guarantees provided
|
75
|
+
- Creator assumes no liability for financial loss
|
76
|
+
- Consult an experienced Generative AI Data Scientist for building your own custom AI Data Science Team
|
77
|
+
- If you want a custom enterprise-grade AI Data Science Team, [send inquiries here](https://www.business-science.io/contact.html).
|
78
|
+
|
79
|
+
By using this software, you agree to use it solely for learning purposes.
|
80
|
+
|
81
|
+
## Table of Contents
|
82
|
+
|
83
|
+
- [Your AI Data Science Team (An Army Of Copilots)](#your-ai-data-science-team-an-army-of-copilots)
|
84
|
+
- [Companies That Want An AI Data Science Team Copilot](#companies-that-want-an-ai-data-science-team-copilot)
|
85
|
+
- [Free Generative AI For Data Scientists Workshop](#free-generative-ai-for-data-scientists-workshop)
|
86
|
+
- [Data Science Agents](#data-science-agents)
|
87
|
+
- [Agents Available Now](#agents-available-now)
|
88
|
+
- [Agents Coming Soon](#agents-coming-soon)
|
89
|
+
- [Disclaimer](#disclaimer)
|
90
|
+
- [Table of Contents](#table-of-contents)
|
91
|
+
- [Installation](#installation)
|
92
|
+
- [Usage](#usage)
|
93
|
+
- [Example 1: Feature Engineering with the Feature Engineering Agent](#example-1-feature-engineering-with-the-feature-engineering-agent)
|
94
|
+
- [Example 2: Cleaning Data with the Data Cleaning Agent](#example-2-cleaning-data-with-the-data-cleaning-agent)
|
95
|
+
- [Contributing](#contributing)
|
96
|
+
- [License](#license)
|
97
|
+
|
98
|
+
## Installation
|
99
|
+
|
100
|
+
``` bash
|
101
|
+
pip install git+https://github.com/business-science/ai-data-science-team.git --upgrade
|
102
|
+
```
|
103
|
+
|
104
|
+
## Usage
|
105
|
+
|
106
|
+
### Example 1: Feature Engineering with the Feature Engineering Agent
|
107
|
+
|
108
|
+
[See the full example here.](/examples/feature_engineering_agent.ipynb)
|
109
|
+
|
110
|
+
``` python
|
111
|
+
feature_engineering_agent = make_feature_engineering_agent(model = llm)
|
112
|
+
|
113
|
+
response = feature_engineering_agent.invoke({
|
114
|
+
"user_instructions": "Make sure to scale and center numeric features",
|
115
|
+
"target_variable": "Churn",
|
116
|
+
"data_raw": df.to_dict(),
|
117
|
+
"max_retries":3,
|
118
|
+
"retry_count":0
|
119
|
+
})
|
120
|
+
```
|
121
|
+
|
122
|
+
``` bash
|
123
|
+
---FEATURE ENGINEERING AGENT----
|
124
|
+
* CREATE FEATURE ENGINEER CODE
|
125
|
+
* EXECUTING AGENT CODE
|
126
|
+
* EXPLAIN AGENT CODE
|
127
|
+
```
|
128
|
+
|
129
|
+
### Example 2: Cleaning Data with the Data Cleaning Agent
|
130
|
+
|
131
|
+
[See the full example here.](/examples/data_cleaning_agent.ipynb)
|
132
|
+
|
133
|
+
``` python
|
134
|
+
data_cleaning_agent = make_data_cleaning_agent(model = llm)
|
135
|
+
|
136
|
+
response = data_cleaning_agent.invoke({
|
137
|
+
"user_instructions": "Don't remove outliers when cleaning the data.",
|
138
|
+
"data_raw": df.to_dict(),
|
139
|
+
"max_retries":3,
|
140
|
+
"retry_count":0
|
141
|
+
})
|
142
|
+
```
|
143
|
+
|
144
|
+
``` bash
|
145
|
+
---DATA CLEANING AGENT----
|
146
|
+
* CREATE DATA CLEANER CODE
|
147
|
+
* EXECUTING AGENT CODE
|
148
|
+
* EXPLAIN AGENT CODE
|
149
|
+
```
|
150
|
+
|
151
|
+
## Contributing
|
152
|
+
|
153
|
+
1. Fork the repository
|
154
|
+
2. Create a feature branch
|
155
|
+
3. Commit your changes
|
156
|
+
4. Push to the branch
|
157
|
+
5. Create a Pull Request
|
158
|
+
|
159
|
+
## License
|
160
|
+
|
161
|
+
This project is licensed under the MIT License. See LICENSE file for details.
|
162
|
+
|
@@ -0,0 +1,19 @@
|
|
1
|
+
ai_data_science_team/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
2
|
+
ai_data_science_team/_version.py,sha256=7tA8TocqCCzLkcB4ptV6bn3k5ni-0TGZvGnVBzmbeIc,26
|
3
|
+
ai_data_science_team/orchestration.py,sha256=xiIFOsrLwPdkSmtme7wNCCGv8XopnMTNElNzlZokL-4,303
|
4
|
+
ai_data_science_team/agents/__init__.py,sha256=DtwQgyeG3Q4rQ-NrMbva-jshVQyULaWW1RrnETQGZOY,270
|
5
|
+
ai_data_science_team/agents/data_cleaning_agent.py,sha256=0K-CgngGjamRk_QzMqNkplrI-ddCbtruQ7kjGrsRIN8,14390
|
6
|
+
ai_data_science_team/agents/data_wrangling_agent.py,sha256=uQBJ8vQwrXubQgaI9_UoNZnVQjIEBUOh3dTmNdg326k,14581
|
7
|
+
ai_data_science_team/agents/feature_engineering_agent.py,sha256=QEqXTsfjllUj4Wgsw4nNGUT6r9Y6q629ZNgqGy3Dbbk,15921
|
8
|
+
ai_data_science_team/templates/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
9
|
+
ai_data_science_team/templates/agent_templates.py,sha256=gT48Pq9KlrrrF0yigodGl_BdptmowTJ2rEWUqh7g5E0,15410
|
10
|
+
ai_data_science_team/tools/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
11
|
+
ai_data_science_team/tools/data_analysis.py,sha256=V7e6_fZA01mosFf5VcLwBcpiMVf7fClZMjTrj-egK-o,3715
|
12
|
+
ai_data_science_team/tools/logging.py,sha256=EU5EMg4Y0-Yhqf1vAEFg0eRvSTx8uF0LTOAKss8-T2M,2073
|
13
|
+
ai_data_science_team/tools/parsers.py,sha256=BAi-fJT7BBt9nRS3w5n9LDTsu7JAJsH8CAI9-Qf7jCs,2086
|
14
|
+
ai_data_science_team/tools/regex.py,sha256=KTH2SXPJT8Tzmj7CufyeET-FbA9BMhRzFlPKr4Tan3g,2320
|
15
|
+
ai_data_science_team-0.0.0.9005.dist-info/LICENSE,sha256=Xif0IRLdd2HGLATxV2EVp91aSY6KOuacRr_6BorKGzA,1084
|
16
|
+
ai_data_science_team-0.0.0.9005.dist-info/METADATA,sha256=PC6rJR965hPu02LtZrzHICkd3QeWzh2A35axTLjE9hM,5840
|
17
|
+
ai_data_science_team-0.0.0.9005.dist-info/WHEEL,sha256=PZUExdf71Ui_so67QXpySuHtCi3-J3wvF4ORK6k_S8U,91
|
18
|
+
ai_data_science_team-0.0.0.9005.dist-info/top_level.txt,sha256=CnoMgOphCoAdGTLueWdCVByVyjwOubaGiTB1lchdy4M,21
|
19
|
+
ai_data_science_team-0.0.0.9005.dist-info/RECORD,,
|
ai_data_science_team/agents.py
DELETED
@@ -1,325 +0,0 @@
|
|
1
|
-
# BUSINESS SCIENCE UNIVERSITY
|
2
|
-
# AI DATA SCIENCE TEAM
|
3
|
-
# ***
|
4
|
-
# Agents
|
5
|
-
# ai_data_science_team/agents.py
|
6
|
-
|
7
|
-
# Libraries
|
8
|
-
from typing import TypedDict, Annotated, Sequence
|
9
|
-
import operator
|
10
|
-
|
11
|
-
from langchain.prompts import PromptTemplate
|
12
|
-
from langchain_core.messages import BaseMessage
|
13
|
-
from langgraph.graph import END, StateGraph
|
14
|
-
|
15
|
-
import os
|
16
|
-
import io
|
17
|
-
import pandas as pd
|
18
|
-
|
19
|
-
from ai_data_science_team.templates.agent_templates import execute_agent_code_on_data, fix_agent_code, explain_agent_code
|
20
|
-
from ai_data_science_team.tools.parsers import PythonOutputParser
|
21
|
-
|
22
|
-
# Setup
|
23
|
-
|
24
|
-
LOG_PATH = os.path.join(os.getcwd(), "logs/")
|
25
|
-
|
26
|
-
|
27
|
-
# * Data Cleaning Agent
|
28
|
-
|
29
|
-
def data_cleaning_agent(model, log=False, log_path=None):
|
30
|
-
"""
|
31
|
-
Creates a data cleaning agent that can be run on a dataset. The agent can be used to clean a dataset in a variety of
|
32
|
-
ways, such as removing columns with more than 40% missing values, imputing missing
|
33
|
-
values with the mean of the column if the column is numeric, or imputing missing
|
34
|
-
values with the mode of the column if the column is categorical.
|
35
|
-
The agent takes in a dataset and some user instructions, and outputs a python
|
36
|
-
function that can be used to clean the dataset. The agent also logs the code
|
37
|
-
generated and any errors that occur.
|
38
|
-
|
39
|
-
Parameters
|
40
|
-
----------
|
41
|
-
model : langchain.llms.base.LLM
|
42
|
-
The language model to use to generate code.
|
43
|
-
log : bool, optional
|
44
|
-
Whether or not to log the code generated and any errors that occur.
|
45
|
-
Defaults to False.
|
46
|
-
log_path : str, optional
|
47
|
-
The path to the directory where the log files should be stored. Defaults to
|
48
|
-
"logs/".
|
49
|
-
|
50
|
-
Examples
|
51
|
-
-------
|
52
|
-
``` python
|
53
|
-
import pandas as pd
|
54
|
-
from langchain_openai import ChatOpenAI
|
55
|
-
from ai_data_science_team.agents import data_cleaning_agent
|
56
|
-
|
57
|
-
llm = ChatOpenAI(model = "gpt-4o-mini")
|
58
|
-
|
59
|
-
data_cleaning_agent = data_cleaning_agent(llm)
|
60
|
-
|
61
|
-
df = pd.read_csv("https://raw.githubusercontent.com/business-science/ai-data-science-team/refs/heads/master/data/churn_data.csv")
|
62
|
-
|
63
|
-
response = data_cleaning_agent.invoke({
|
64
|
-
"user_instructions": "Don't remove outliers when cleaning the data.",
|
65
|
-
"data_raw": df.to_dict(),
|
66
|
-
"max_retries":3,
|
67
|
-
"retry_count":0
|
68
|
-
})
|
69
|
-
|
70
|
-
pd.DataFrame(response['data_cleaned'])
|
71
|
-
```
|
72
|
-
|
73
|
-
Returns
|
74
|
-
-------
|
75
|
-
app : langchain.graphs.StateGraph
|
76
|
-
The data cleaning agent as a state graph.
|
77
|
-
"""
|
78
|
-
llm = model
|
79
|
-
|
80
|
-
# Setup Log Directory
|
81
|
-
if log:
|
82
|
-
if log_path is None:
|
83
|
-
log_path = LOG_PATH
|
84
|
-
if not os.path.exists(log_path):
|
85
|
-
os.makedirs(log_path)
|
86
|
-
|
87
|
-
# Define GraphState for the router
|
88
|
-
class GraphState(TypedDict):
|
89
|
-
messages: Annotated[Sequence[BaseMessage], operator.add]
|
90
|
-
user_instructions: str
|
91
|
-
data_raw: dict
|
92
|
-
data_cleaner_function: str
|
93
|
-
data_cleaner_error: str
|
94
|
-
data_cleaned: dict
|
95
|
-
max_retries: int
|
96
|
-
retry_count: int
|
97
|
-
|
98
|
-
|
99
|
-
def create_data_cleaner_code(state: GraphState):
|
100
|
-
print("---DATA CLEANING AGENT----")
|
101
|
-
print(" * CREATE DATA CLEANER CODE")
|
102
|
-
|
103
|
-
data_cleaning_prompt = PromptTemplate(
|
104
|
-
template="""
|
105
|
-
You are a Data Cleaning Agent. Your job is to create a data_cleaner() function to that can be run on the data provided.
|
106
|
-
|
107
|
-
Things that should be considered in the data summary function:
|
108
|
-
|
109
|
-
* Removing columns if more than 40 percent of the data is missing
|
110
|
-
* Imputing missing values with the mean of the column if the column is numeric
|
111
|
-
* Imputing missing values with the mode of the column if the column is categorical
|
112
|
-
* Converting columns to the correct data type
|
113
|
-
* Removing duplicate rows
|
114
|
-
* Removing rows with missing values
|
115
|
-
* Removing rows with extreme outliers (3X the interquartile range)
|
116
|
-
|
117
|
-
Make sure to take into account any additional user instructions that may negate some of these steps or add new steps. Include comments in your code to explain your reasoning for each step. Include comments if something is not done because a user requested. Include comments if something is done because a user requested.
|
118
|
-
|
119
|
-
User instructions:
|
120
|
-
{user_instructions}
|
121
|
-
|
122
|
-
Return Python code in ```python ``` format with a single function definition, data_cleaner(data_raw), that incldues all imports inside the function.
|
123
|
-
|
124
|
-
You can use Pandas, Numpy, and Scikit Learn libraries to clean the data.
|
125
|
-
|
126
|
-
Sample Data (first 100 rows):
|
127
|
-
{data_head}
|
128
|
-
|
129
|
-
Data Description:
|
130
|
-
{data_description}
|
131
|
-
|
132
|
-
Data Info:
|
133
|
-
{data_info}
|
134
|
-
|
135
|
-
Return code to provide the data cleaning function:
|
136
|
-
|
137
|
-
def data_cleaner(data_raw):
|
138
|
-
import pandas as pd
|
139
|
-
import numpy as np
|
140
|
-
...
|
141
|
-
return data_cleaner
|
142
|
-
|
143
|
-
Best Practices and Error Preventions:
|
144
|
-
|
145
|
-
Always ensure that when assigning the output of fit_transform() from SimpleImputer to a Pandas DataFrame column, you call .ravel() or flatten the array, because fit_transform() returns a 2D array while a DataFrame column is 1D.
|
146
|
-
|
147
|
-
""",
|
148
|
-
input_variables=["user_instructions","data_head", "data_description", "data_info"]
|
149
|
-
)
|
150
|
-
|
151
|
-
data_cleaning_agent = data_cleaning_prompt | llm | PythonOutputParser()
|
152
|
-
|
153
|
-
data_raw = state.get("data_raw")
|
154
|
-
|
155
|
-
df = pd.DataFrame.from_dict(data_raw)
|
156
|
-
|
157
|
-
buffer = io.StringIO()
|
158
|
-
df.info(buf=buffer)
|
159
|
-
info_text = buffer.getvalue()
|
160
|
-
|
161
|
-
response = data_cleaning_agent.invoke({
|
162
|
-
"user_instructions": state.get("user_instructions"),
|
163
|
-
"data_head": df.head().to_string(),
|
164
|
-
"data_description": df.describe().to_string(),
|
165
|
-
"data_info": info_text
|
166
|
-
})
|
167
|
-
|
168
|
-
# For logging: store the code generated:
|
169
|
-
if log:
|
170
|
-
with open(log_path + 'data_cleaner.py', 'w') as file:
|
171
|
-
file.write(response)
|
172
|
-
|
173
|
-
return {"data_cleaner_function" : response}
|
174
|
-
|
175
|
-
def execute_data_cleaner_code(state):
|
176
|
-
return execute_agent_code_on_data(
|
177
|
-
state=state,
|
178
|
-
data_key="data_raw",
|
179
|
-
result_key="data_cleaned",
|
180
|
-
error_key="data_cleaner_error",
|
181
|
-
code_snippet_key="data_cleaner_function",
|
182
|
-
agent_function_name="data_cleaner",
|
183
|
-
pre_processing=lambda data: pd.DataFrame.from_dict(data),
|
184
|
-
post_processing=lambda df: df.to_dict(),
|
185
|
-
error_message_prefix="An error occurred during data cleaning: "
|
186
|
-
)
|
187
|
-
|
188
|
-
def fix_data_cleaner_code(state: GraphState):
|
189
|
-
data_cleaner_prompt = """
|
190
|
-
You are a Data Cleaning Agent. Your job is to create a data_cleaner() function that can be run on the data provided. The function is currently broken and needs to be fixed.
|
191
|
-
|
192
|
-
Make sure to only return the function definition for data_cleaner().
|
193
|
-
|
194
|
-
Return Python code in ```python``` format with a single function definition, data_cleaner(data_raw), that includes all imports inside the function.
|
195
|
-
|
196
|
-
This is the broken code (please fix):
|
197
|
-
{code_snippet}
|
198
|
-
|
199
|
-
Last Known Error:
|
200
|
-
{error}
|
201
|
-
"""
|
202
|
-
|
203
|
-
return fix_agent_code(
|
204
|
-
state=state,
|
205
|
-
code_snippet_key="data_cleaner_function",
|
206
|
-
error_key="data_cleaner_error",
|
207
|
-
llm=llm,
|
208
|
-
prompt_template=data_cleaner_prompt,
|
209
|
-
log=True,
|
210
|
-
log_path="logs/",
|
211
|
-
log_file_name="data_cleaner.py"
|
212
|
-
)
|
213
|
-
|
214
|
-
def explain_data_cleaner_code(state: GraphState):
|
215
|
-
return explain_agent_code(
|
216
|
-
state=state,
|
217
|
-
code_snippet_key="data_cleaner_function",
|
218
|
-
result_key="messages",
|
219
|
-
error_key="data_cleaner_error",
|
220
|
-
llm=llm,
|
221
|
-
explanation_prompt_template="""
|
222
|
-
Explain the data cleaning steps that the data cleaning agent performed in this function.
|
223
|
-
Keep the summary succinct and to the point.\n\n# Data Cleaning Agent:\n\n{code}
|
224
|
-
""",
|
225
|
-
success_prefix="# Data Cleaning Agent:\n\n ",
|
226
|
-
error_message="The Data Cleaning Agent encountered an error during data cleaning. Data could not be explained."
|
227
|
-
)
|
228
|
-
|
229
|
-
|
230
|
-
workflow = StateGraph(GraphState)
|
231
|
-
|
232
|
-
workflow.add_node("create_data_cleaner_code", create_data_cleaner_code)
|
233
|
-
workflow.add_node("execute_data_cleaner_code", execute_data_cleaner_code)
|
234
|
-
workflow.add_node("fix_data_cleaner_code", fix_data_cleaner_code)
|
235
|
-
workflow.add_node("explain_data_cleaner_code", explain_data_cleaner_code)
|
236
|
-
|
237
|
-
workflow.set_entry_point("create_data_cleaner_code")
|
238
|
-
workflow.add_edge("create_data_cleaner_code", "execute_data_cleaner_code")
|
239
|
-
|
240
|
-
workflow.add_conditional_edges(
|
241
|
-
"execute_data_cleaner_code",
|
242
|
-
lambda state: "fix_code"
|
243
|
-
if (state.get("data_cleaner_error") is not None
|
244
|
-
and state.get("retry_count") is not None
|
245
|
-
and state.get("max_retries") is not None
|
246
|
-
and state.get("retry_count") < state.get("max_retries"))
|
247
|
-
else "explain_code",
|
248
|
-
{"fix_code": "fix_data_cleaner_code", "explain_code": "explain_data_cleaner_code"},
|
249
|
-
)
|
250
|
-
|
251
|
-
workflow.add_edge("fix_data_cleaner_code", "execute_data_cleaner_code")
|
252
|
-
workflow.add_edge("explain_data_cleaner_code", END)
|
253
|
-
|
254
|
-
app = workflow.compile()
|
255
|
-
|
256
|
-
return app
|
257
|
-
|
258
|
-
# # * Data Summary Agent
|
259
|
-
|
260
|
-
# def data_summary_agent(model, log=True, log_path=None):
|
261
|
-
|
262
|
-
# # Setup Log Directory
|
263
|
-
# if log:
|
264
|
-
# if log_path is None:
|
265
|
-
# log_path = LOG_PATH
|
266
|
-
# if not os.path.exists(log_path):
|
267
|
-
# os.makedirs(log_path)
|
268
|
-
|
269
|
-
# llm = model
|
270
|
-
|
271
|
-
# data_summary_prompt = PromptTemplate(
|
272
|
-
# template="""
|
273
|
-
# You are a Data Summary Agent. Your job is to summarize a dataset.
|
274
|
-
|
275
|
-
# Things that should be considered in the data summary function:
|
276
|
-
|
277
|
-
# * How many missing values
|
278
|
-
# * How many unique values
|
279
|
-
# * How many rows
|
280
|
-
# * How many columns
|
281
|
-
# * What data types are present
|
282
|
-
# * What the data looks like
|
283
|
-
# * What column types are present
|
284
|
-
# * What is the distribution of the data
|
285
|
-
# * What is the correlation between the data
|
286
|
-
|
287
|
-
# Make sure to take into account any additional user instructions that may negate some of these steps or add new steps.
|
288
|
-
|
289
|
-
# User instructions:
|
290
|
-
# {user_instructions}
|
291
|
-
|
292
|
-
# Return Python code in ```python ``` format with a single function definition, data_sumary(data), that incldues all imports inside the function.
|
293
|
-
|
294
|
-
# You can use Pandas, Numpy, and Scikit Learn libraries to summarize the data.
|
295
|
-
|
296
|
-
# Sample Data (first 100 rows):
|
297
|
-
# {data_head}
|
298
|
-
|
299
|
-
# Data Description:
|
300
|
-
# {data_description}
|
301
|
-
|
302
|
-
# Data Info:
|
303
|
-
# {data_info}
|
304
|
-
|
305
|
-
# Return code to provide the data cleaning function:
|
306
|
-
|
307
|
-
# def data_summary(data):
|
308
|
-
# import pandas as pd
|
309
|
-
# import numpy as np
|
310
|
-
# ...
|
311
|
-
# return {
|
312
|
-
# 'data_summary': ...,
|
313
|
-
# 'data_correlation': ...
|
314
|
-
# [INSERT MORE KEYS HERE],
|
315
|
-
# }
|
316
|
-
|
317
|
-
# """,
|
318
|
-
# input_variables=["user_instructions","data_head", "data_description", "data_info"]
|
319
|
-
# )
|
320
|
-
|
321
|
-
# data_summary_agent = data_summary_prompt | llm | PythonOutputParser()
|
322
|
-
|
323
|
-
|
324
|
-
|
325
|
-
# return 1
|