A set of AI tools for working with Cognite Data Fusion in Python.
Project description
cognite-ai
A set of AI tools for working with CDF (Cognite Data Fusion) in Python, including vector stores and intelligent data manipulation features leveraging large language models (LLMs).
Installation
This package is intended to be used in Cognite's Jupyter notebook and Streamlit. To get started, install the package using:
%pip install cognite-ai
MemoryVectorStore
The MemoryVectorStore
allows you to store and query vector embeddings created from text, enabling use cases where the number of vectors is relatively small.
Example
You can create vectors from text (either as individual strings or as a list) and query them:
from cognite.ai import MemoryVectorStore
from cognite.client import CogniteClient
client = CogniteClient()
# Create a MemoryVectorStore instance
vector_store = MemoryVectorStore(client)
# Store text as vectors
vector_store.store_text("The compressor in unit 7B requires maintenance next week.")
vector_store.store_text("Pump 5A has shown signs of decreased efficiency.")
vector_store.store_text("Unit 9 is operating at optimal capacity.")
# Query vector store
vector_store.query_text("Which units require maintenance?")
Smart Data Tools
With cognite-ai
, you can enhance your data workflows by integrating LLMs for intuitive querying and manipulation of data frames. The module is built on top of PandasAI and adds Cognite-specific features.
The Smart Data Tools come in three components:
Pandas Smart DataFrame Pandas Smart DataLake Pandas AI Agent
1. Pandas Smart DataFrame
SmartDataframe
enables you to chat with individual data frames, using LLMs to query, summarize, and analyze your data conversationally.
Example
from cognite.ai import load_pandasai
from cognite.client import CogniteClient
import pandas as pd
# Load the necessary classes
client = CogniteClient()
SmartDataframe, SmartDatalake, Agent = await load_pandasai()
# Create demo data
workorders_df = pd.DataFrame({
"workorder_id": ["WO001", "WO002", "WO003", "WO004", "WO005"],
"description": [
"Replace filter in compressor unit 3A",
"Inspect and lubricate pump 5B",
"Check pressure valve in unit 7C",
"Repair leak in pipeline 4D",
"Test emergency shutdown system"
],
"priority": ["High", "Medium", "High", "Low", "Medium"]
})
# Create a SmartDataframe object
s_workorders_df = SmartDataframe(workorders_df, cognite_client=client)
# Chat with the dataframe
s_workorders_df.chat('Which 5 work orders are the most critical based on priority?')
Customizing LLM Parameters
You can configure the LLM parameters to control aspects like model selection and temperature.
params = {
"model": "gpt-35-turbo",
"temperature": 0.5
}
s_workorders_df = SmartDataframe(workorders_df, cognite_client=client, params=params)
2. Pandas Smart DataLake
SmartDatalake
allows you to combine and query multiple data frames simultaneously, treating them as a unified data lake.
Example
from cognite.ai import load_pandasai
from cognite.client import CogniteClient
import pandas as pd
# Load the necessary classes
client = CogniteClient()
SmartDataframe, SmartDatalake, Agent = await load_pandasai()
# Create demo data
workorders_df = pd.DataFrame({
"workorder_id": ["WO001", "WO002", "WO003"],
"asset_id": ["A1", "A2", "A3"],
"description": ["Replace filter", "Inspect pump", "Check valve"]
})
workitems_df = pd.DataFrame({
"workitem_id": ["WI001", "WI002", "WI003"],
"workorder_id": ["WO001", "WO002", "WO003"],
"task": ["Filter replacement", "Pump inspection", "Valve check"]
})
assets_df = pd.DataFrame({
"asset_id": ["A1", "A2", "A3"],
"name": ["Compressor 3A", "Pump 5B", "Valve 7C"]
})
# Combine them into a smart lake
smart_lake_df = SmartDatalake([workorders_df, workitems_df, assets_df], cognite_client=client)
# Chat with the unified data lake
smart_lake_df.chat("Which assets have the most work orders associated with them?")
3. Pandas AI Agent
The Agent
provides conversational querying capabilities across a single data frame, allowing you to have follow up questions.
Example
from cognite.ai import load_pandasai
from cognite.client import CogniteClient
import pandas as pd
# Load the necessary classes
client = CogniteClient()
SmartDataframe, SmartDatalake, Agent = await load_pandasai()
# Create example data
sensor_readings_df = pd.DataFrame({
"sensor_id": ["A1", "A2", "A3", "A4", "A5"],
"temperature": [75, 80, 72, 78, 69],
"pressure": [30, 35, 33, 31, 29],
"status": ["Normal", "Warning", "Normal", "Warning", "Normal"]
})
# Create an Agent for the dataframe
agent = Agent(sensor_readings_df, cognite_client=client)
# Ask a question
print(agent.chat("Which sensors are showing a warning status?"))
Contributing
This package exists mainly to provide a in memory vector store
in addition to getting around the install problems
a user gets in Pyodide when installing pandasai
due to
dependencies that are not pure Python 3 wheels.
The current development cycle is not great, but consists of copying the contents of the source code in this package into e.g. a Jupyter notebook in Fusion to verify that everything works there.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file cognite_ai-0.6.0.tar.gz
.
File metadata
- Download URL: cognite_ai-0.6.0.tar.gz
- Upload date:
- Size: 9.0 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/5.1.1 CPython/3.12.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 5d02e9f701fbe7c26ae64393ad85aa058a2c2dc0f9f6ad86d7391273357423c7 |
|
MD5 | c81bc154bcd942404303f9691facdfa0 |
|
BLAKE2b-256 | becdfe954f27183f890071295004b96df4d9d965a786480123413cdf60a90278 |
Provenance
File details
Details for the file cognite_ai-0.6.0-py3-none-any.whl
.
File metadata
- Download URL: cognite_ai-0.6.0-py3-none-any.whl
- Upload date:
- Size: 9.2 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/5.1.1 CPython/3.12.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | cddbb3e97e17942ce5aade57996abf0112bf85c843fcf54dbaffe0d565c345d9 |
|
MD5 | 26507e57be7f588955fe8c9153902664 |
|
BLAKE2b-256 | 2a5df7e06b51f470c6fa2e96c7dba5ee0858e81564f37665bd2bc07ab8df5609 |