Spaces:
Sleeping
Sleeping
File size: 46,536 Bytes
754cec9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 |
# src/agents.py
import base64
import os
import logging
import functools
import subprocess
import sys
from io import StringIO
from typing import List, Annotated, Sequence, TypedDict
import operator
from langchain.agents import AgentExecutor, create_openai_tools_agent
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.output_parsers.openai_functions import JsonOutputFunctionsParser
from langchain_core.messages import BaseMessage, HumanMessage
from langchain_experimental.tools import PythonREPLTool
from langchain_openai import ChatOpenAI, OpenAI
from langchain_core.messages import AIMessage
from langchain_experimental.agents.agent_toolkits.pandas.base import create_pandas_dataframe_agent
from pydantic import BaseModel, Field, PrivateAttr
from langchain_core.tools import StructuredTool
from langchain.agents.format_scratchpad.openai_tools import (
format_to_openai_tool_messages,
)
from langchain.agents.output_parsers.openai_tools import OpenAIToolsAgentOutputParser
from langgraph.graph import StateGraph, END
from langchain.agents.agent_types import AgentType
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import Chroma
from typing import Any
import streamlit as st
import pandas as pd
import matplotlib.pyplot as plt
from openai import OpenAI
#sys.path
# Get the absolute path of the current file (agents.py)
current_dir = os.path.dirname(os.path.abspath(__file__))
# Get the parent directory
parent_dir = os.path.abspath(os.path.join(current_dir, '..'))
# Add the parent directory to sys.path
if parent_dir not in sys.path:
sys.path.insert(0, parent_dir)
# Import custom modules
from .search.search_pg_default import pg_search_default
from .search.publication_qa_tool import answer_publication_questions, PublicationQAArgs
from .plotting_tools.hard_agent import plot_master_track_map
from .plotting_tools.oceanographer_tools import plot_ts_diagram
from .prompts import Prompts
from .utils import generate_unique_image_path
from .config import API_KEY
# 1. Search Agent and Tools
class CustomPythonREPLTool(PythonREPLTool):
_datasets: dict = PrivateAttr()
def __init__(self, datasets, **kwargs):
"""
Custom Python REPL tool that injects dataset variables and logs plot generation.
:param datasets: Dictionary { "dataset_1": <DataFrame>, "dataset_2": <DataFrame>, ... }
"""
super().__init__(**kwargs)
self._datasets = datasets
def _run(self, query: str, **kwargs) -> Any:
"""
Execute the user-provided Python code in a local context containing:
- st (Streamlit)
- plt (Matplotlib Pyplot)
- pd (Pandas)
- All loaded dataset variables (self._datasets)
- A dynamically generated plot_path
If a figure is saved to plot_path, a "plot_generated" event will be logged in session_state["execution_history"].
"""
import streamlit as st
import matplotlib.pyplot as plt
import pandas as pd
import logging
from io import StringIO
from src.utils import log_history_event, generate_unique_image_path
# Prepare local context with necessary packages
local_context = {"st": st, "plt": plt, "pd": pd}
# Inject the user’s datasets under the specified variable names (e.g. dataset_1, dataset_2, etc.)
local_context.update(self._datasets)
# Generate a unique file path for the plot (plot_path)
plot_path = generate_unique_image_path()
local_context['plot_path'] = plot_path
# Redirect stdout so we can capture any output from exec(...)
old_stdout = sys.stdout
redirected_output = StringIO()
sys.stdout = redirected_output
try:
# Execute user code
exec(query, local_context)
output = redirected_output.getvalue()
except ModuleNotFoundError as e:
missing_module = e.name
logging.warning(f"Module '{missing_module}' not found during code execution.")
return {
"error": "ModuleNotFoundError",
"missing_module": missing_module,
"message": f"The Python module '{missing_module}' is not installed."
}
except Exception as e:
logging.error(f"Error during code execution: {e}")
return {
"error": "ExecutionError",
"message": str(e)
}
finally:
# Restore stdout
sys.stdout = old_stdout
# Check if a plot was actually saved to plot_path
plot_generated = False
if os.path.exists(plot_path):
st.session_state.saved_plot_path = plot_path
st.session_state.plot_image = plot_path
st.session_state.new_plot_path = plot_path
plot_generated = True
if plot_generated:
status_message = f"Plot generated = True. Saved at: {plot_path}"
logging.info(status_message)
st.session_state.plot_generated_status = status_message
from src.utils import log_history_event
log_history_event(
st.session_state,
"plot_generated",
{
"plot_path": plot_path.replace("sandbox:", ""), # Remove sandbox prefix
"agent": "VisualizationAgent",
"description": "Python_REPL Generated Plot",
"content": query # Store the actual code used
}
)
return {
"result": f"Execution completed. Plot saved at: {plot_path if plot_generated else 'No plot generated'}",
"output": output,
"plot_images": [plot_path] if plot_generated else []
}
def search_pg_datasets_tool(query):
global prompt_search
datasets_info = pg_search_default(query)
logging.debug("Datasets info: %s", datasets_info)
if not datasets_info.empty:
st.session_state.datasets_info = datasets_info
st.session_state.messages_search.append({
"role": "assistant",
"content": f"**Search query:** {query}"
})
# Pass the table as JSON (you can use orient="split" or the default, as long as your UI can parse it)
st.session_state.messages_search.append({
"role": "assistant",
"content": "**Datasets Information:**",
"table": datasets_info.to_json(orient="split")
})
# Optionally, build a detailed description string for the prompt:
datasets_description = ""
for i, row in datasets_info.iterrows():
datasets_description += (
f"Dataset {i + 1}:\n"
f"Name: {row['Name']}\n"
f"Description: {row['Short Description']}\n"
f"Parameters: {row['Parameters']}\n\n"
)
prompt_search = (
f"The user has provided the following query: {query}\n"
f"Available datasets:\n{datasets_description}\n"
"Please identify the top two datasets that best match the user's query and explain why they are the most relevant. "
"Do not suggest datasets without values in the Parameters field, because they cannot be directly downloaded.\n"
"Respond using the following schema:\n"
"{dataset name}\n{reason why relevant}\n{propose some short analysis and further questions to answer}"
)
return datasets_info, prompt_search
def create_search_agent(datasets_info=None):
model_name = st.session_state.get("model_name", "gpt-3.5-turbo")
if model_name == "o3-mini":
llm = ChatOpenAI(api_key=API_KEY, model_name=model_name)
else:
llm = ChatOpenAI(api_key=API_KEY, model_name=model_name)
# Generate dataset description string
datasets_description = ""
if datasets_info is not None:
for i, row in datasets_info.iterrows():
datasets_description += f"Dataset {i + 1}:\nName: {row['Name']}\nDOI: {row['DOI']}\nDescription: {row['Short Description']}\nParameters: {row['Parameters']}\n\n"
prompt = ChatPromptTemplate.from_messages(
[
("system",
f"You are a powerful assistant primarily designed to search and retrieve datasets from PANGAEA. Your main goal is to help users find relevant datasets using the search_pg_datasets tool. When a user asks about datasets, always use this tool first to provide the most up-to-date and accurate information.\n\n"
#f"Here are some datasets returned from the search:\n{datasets_description}"
"In addition to dataset searches, you have a secondary capability to answer questions about publications related to specific datasets (or in other words what was published based on this dataset). If a user explicitly asks about publications or research findings based on a particular dataset, you can use the answer_publication_questions tool. For example, you can handle queries like 'What was published based on this dataset?' or 'What were the main conclusions from the research using this dataset?'\n\n"
"Remember:\n"
"1. Prioritize dataset searches using the search_pg_datasets tool. Make sure that the query you pass to the tool is rephrased so that elastic search gives the best match. Also try not to include words like 'search' and etc. in the search query.\n"
"2. Only use the answer_publication_questions tool when the user specifically asks about publications or research findings related to a dataset they've already identified. Please make sure that you correctly pass the doi to the tool. It should be doi retrieved after the search (user will point out which dataset it interested in). DO NOT GENERATE DOI ON THIS STEP OUT OF YOUR MIND! JUST TAKE WHAT WAS GIVEN WITH SYSTEM PROMPT.\n"
"3. If needed, ask the user to clarify which dataset they're referring to before using the publication tool.\n\n"
"Strive to provide accurate, helpful, and concise responses to user queries."
),
("user", "{input}"),
MessagesPlaceholder(variable_name="chat_history"),
MessagesPlaceholder(variable_name="agent_scratchpad"),
]
)
search_tool = StructuredTool.from_function(
func=search_pg_datasets_tool,
name="search_pg_datasets",
description="List datasets from PANGAEA based on a query"
)
publication_qa_tool = StructuredTool.from_function(
func=answer_publication_questions,
name="answer_publication_questions",
description="A tool to answer questions about articles published from this dataset. This will be a journal article for which you should provide the tool with an already structured question about what the user wants. The input should be the DOI of the dataset (e.g. 'https://doi.org/10.1594/PANGAEA.xxxxxx') and the question. The question should be reworded to specifically send it to RAG. E.g. the hypothetical user's question 'Are there any related articles to the first dataset? If so what these articles are about?' will be re-worded for this tool as 'What is this article is about?'",
args_schema=PublicationQAArgs
)
tools = [search_tool, publication_qa_tool]
llm_with_tools = llm.bind_tools(tools)
agent = (
{
"input": lambda x: x["input"],
"chat_history": lambda x: x.get("chat_history", []),
"agent_scratchpad": lambda x: format_to_openai_tool_messages(x["intermediate_steps"]),
}
| prompt
| llm_with_tools
| OpenAIToolsAgentOutputParser()
)
return AgentExecutor(agent=agent, tools=tools, verbose=True, max_iterations=5)
# 2. Visualization and Oceanography Tools
class PlotMasterTrackMapArgs(BaseModel):
dataset_var: str = Field(description="The variable name of the dataset to use (e.g., 'dataset_1', 'dataset_2').")
main_title: str = Field(description="The main title for the plot.")
lat_col: str = Field(description="Name of the latitude column.")
lon_col: str = Field(description="Name of the longitude column.")
date_col: str = Field(description="Name of the date/time column.")
class TSPlotToolArgs(BaseModel):
dataset_var: str = Field(description="The variable name of the dataset to use (e.g., 'dataset_1', 'dataset_2').")
main_title: str = Field(description="The main title for the plot.")
temperature_col: str = Field(description="Name of the temperature column.")
salinity_col: str = Field(description="Name of the salinity column.")
# 3. Agent Creation Functions
def create_pandas_agent(user_query, datasets_info):
if st.session_state.model_name == "o3-mini":
llm = ChatOpenAI(api_key=API_KEY, model_name=st.session_state.model_name)
else:
llm = ChatOpenAI(api_key=API_KEY, model_name=st.session_state.model_name)
# Assign unique variable names to each dataframe and collect dataframes
dataset_variables = []
dataframes = []
datasets_text = "" # Initialize datasets_text
for i, info in enumerate(datasets_info, 1): # Start enumeration at 1
var_name = f"df{i}" # Consistently name as df1, df2, etc.
dataframes.append(info['dataset']) # Collect dataframes into a list
dataset_variables.append(var_name)
# Build datasets_text
datasets_text += (
f"Dataset {i}:\n" # Adjust index to match variable naming
f"Variable Name: {var_name}\n"
f"Name: {info['name']}\n"
f"Description: {info['description']}\n"
f"Head of DataFrame (use it only as an example):\n"
f"{info['df_head']}\n\n"
)
# Create a custom system prompt that includes information about each dataframe
system_prompt = Prompts.generate_pandas_agent_system_prompt(user_query, datasets_text, dataset_variables)
# Create a ChatPromptTemplate with the system prompt
chat_prompt = ChatPromptTemplate.from_messages(
[
("system", system_prompt),
("user", "{input}"),
MessagesPlaceholder(variable_name="chat_history"),
MessagesPlaceholder(variable_name="agent_scratchpad"),
]
)
# Create the pandas dataframe agent with the list of dataframes and the chat prompt
agent_pandas = create_pandas_dataframe_agent(
llm=llm,
df=dataframes, # Pass the list of dataframes
agent_type=AgentType.OPENAI_FUNCTIONS,
verbose=True,
return_intermediate_steps=True,
max_iterations=5,
early_stopping_method="generate",
handle_parsing_errors=True,
#prefix=system_prompt,
suffix=system_prompt,
allow_dangerous_code=True,
chat_prompt=chat_prompt
)
return agent_pandas
# Define the function to encode the image
def encode_image(image_path):
with open(image_path, "rb") as image_file:
return base64.b64encode(image_file.read()).decode('utf-8')
def reflect_on_image(image_path: str) -> str:
if not os.path.exists(image_path):
return f"Error: The file {image_path} does not exist."
base64_image = encode_image(image_path)
prompt = """You are a professional reviewer of scientific images. Your task is to provide review and pass it back to the visual creator agent, so that it could improve it. At each step provide idea for improvements (only if necessary). Be sure to be critical and provide a source for improvement. Conduct a quality check of the provided image using the following criteria:
1. Axis and Font Quality: Evaluate the visibility of axes and appropriateness of font size and style. Are the axes clearly visible and labeled? Is the font legible and suitable for the image size?
2. Label Clarity: Assess if labels are well-positioned and not overlapping. Are all labels clearly readable and properly placed?
3. Color Scheme: Analyze the color choices. Is the color scheme appropriate for the data presented? Are the colors distinguishable and not causing visual confusion?
4. Data Representation: Evaluate how well the data is represented. Are data points clearly visible? Is the chosen chart or graph type appropriate for the data?
5. Legend and Scale: Check the presence and clarity of legends and scales. Are they present when necessary and easy to understand?
6. Overall Layout: Assess the overall layout and use of space. Is the image well-balanced and visually appealing?
7. Technical Issues: Identify any technical problems such as pixelation, blurriness, or artifacts that might affect the image quality.
8. Scientific Accuracy: To the best of your ability, comment on whether the image appears scientifically accurate and free from obvious errors or misrepresentations.
9. Check that the figure make sense from an observing human's point of view, for example, if the figure have a variable ‘Depth of water or smth’ it should be on the Y-AXIS and go from surface to depth, so minimum at the top, max depth in the bottom. If there are remarks about these things, severely underestimate the final mark for the figure and force agent to redo the graph, with precise instructions. SUPER IMPORTANT -> IF DEPTH OF WATER OR ANY VERTICAL DIMENSIONS ARE PRESENT, AND THEY ARE ON THE HORIZONTAL X-AXIS, AND NOT ON Y-AXIS, RETURN FIGURE BACK WITH SCORE 1/10, PUNISH SEVERELY FOR THIS! <- SUPER IMPORTANT
Please provide a structured review addressing each of these points. Conclude with an overall assessment of the image quality, highlighting any significant issues or exemplary aspects. Finally, give the image a score out of 10, where 10 is perfect quality and 1 is unusable.
"""
openai_client = OpenAI(api_key=API_KEY)
response = openai_client.chat.completions.create(
model="gpt-4o",
messages=[
{
"role": "user",
"content": [
{"type": "text", "text": prompt},
{
"type": "image_url",
"image_url": {
"url": f"data:image/png;base64,{base64_image}"
}
}
]
}
],
max_tokens=1000
)
return response.choices[0].message.content
# Define the args schema for reflect_on_image
class ReflectOnImageArgs(BaseModel):
image_path: str = Field(description="The path to the image to reflect on.")
# Define the reflect_on_image tool
reflect_tool = StructuredTool.from_function(
func=reflect_on_image,
name="reflect_on_image",
description="A tool to reflect on an image and provide feedback for improvements.",
args_schema=ReflectOnImageArgs
)
#Planning tool
class PlanningToolArgs(BaseModel):
goal: str = Field(
description="A short statement of the user's main objective or question to create a plan for."
)
constraints: List[str] = Field(
default_factory=list,
description="Any constraints or conditions to be respected in the plan (e.g., time or resource constraints)."
)
user_query: str = Field(
default="",
description="The original user query or question that triggered the plan request."
)
datasets_summary: str = Field(
default="",
description="A concise summary of the current datasets or project context that the plan should consider."
)
def planning_tool(
goal: str,
constraints: List[str],
user_query: str,
datasets_summary: str
) -> dict:
"""
A planning function that uses a ChatCompletion to create a step-by-step plan,
referencing the user query, constraints, and dataset info for context.
Returns a dict with at least "messages" so it updates the state in langgraph.
"""
from langchain_openai import ChatOpenAI
from langchain_core.messages import AIMessage, SystemMessage, HumanMessage
# Create a system prompt that instructs the LLM how to create the plan:
system_prompt = (
"You are an advanced 'PlanningTool' that must generate a step-by-step plan. "
"Consider the user’s ultimate goal, the constraints, the original query, and the dataset context. "
"Respond with a thorough but concise plan that can be used by the system to coordinate tasks."
)
# We'll build a user message that includes all relevant info:
# (goal, constraints, user_query, and the dataset summary).
user_message = (
f"Goal: {goal}\n\n"
f"Constraints: {constraints}\n\n"
f"User Query: {user_query}\n\n"
f"Dataset Info:\n{datasets_summary}\n\n"
"Please produce a plan with carefully enumerated steps."
)
# Create an LLM instance
model_name = st.session_state.get("model_name", "gpt-3.5-turbo")
if model_name == "o3-mini":
llm = ChatOpenAI(api_key=API_KEY, model_name=model_name)
else:
llm = ChatOpenAI(api_key=API_KEY, model_name=model_name)
# Construct messages for the chat
messages = [
SystemMessage(content=system_prompt),
HumanMessage(content=user_message)
]
# Call the LLM
response = llm(messages)
# The text of the plan is in response.content
final_plan_text = response.content
# Return a dictionary that merges into state["messages"]
# (this is how the graph update won't fail with InvalidUpdateError)
return {
"messages": [
AIMessage(content=final_plan_text, name="Planner")
]
}
def install_package(package_name: str, pip_options: str = ""):
#ALLOWED_PACKAGES = {"matplotlib", "seaborn", "plotly", "pandas", "numpy", "gsw", "scipy"}
#if package_name not in ALLOWED_PACKAGES:
# return f"Installation of package '{package_name}' is not allowed."
try:
command = [sys.executable, '-m', 'pip', 'install'] + pip_options.split() + [package_name]
subprocess.check_call(command)
return f"Package '{package_name}' installed successfully."
except Exception as e:
return f"Failed to install package '{package_name}': {e}"
# Define the args schema for install_package
class InstallPackageArgs(BaseModel):
package_name: str = Field(description="The name of the package to install.")
pip_options: str = Field(default="", description="Additional pip options (e.g., '--force-reinstall').")
# Create the install_package_tool
install_package_tool = StructuredTool.from_function(
func=install_package,
name="install_package",
description="Installs a Python package using pip. Use this tool if you encounter a ModuleNotFoundError or need a package that's not installed.",
args_schema=InstallPackageArgs
)
def get_example_of_visualizations(query: str) -> str:
"""
Retrieves example visualizations related to the query.
Parameters:
- query (str): The user's query about plotting.
Returns:
- str: The content of the most relevant example file.
"""
# Initialize embeddings
#api_key = st.secrets["general"]["openai_api_key"]
embeddings = OpenAIEmbeddings(api_key=API_KEY)
# Load the existing vector store
vector_store = Chroma(
collection_name="example_collection",
embedding_function=embeddings,
persist_directory=os.path.join('data', 'examples_database', 'chroma_langchain_notebooks')
)
# Perform a similarity search
results = vector_store.similarity_search_with_score(query, k=1)
# Extract the most relevant document
doc, score = results[0]
# Construct the full path to the txt file
file_name = doc.metadata['source'].lstrip('./')
full_path = os.path.join('data', 'examples_database', file_name)
# Read and return the content of the txt file
try:
with open(full_path, 'r', encoding='utf-8') as file:
content = file.read()
return content
except Exception as e:
logging.error(f"An error occurred while reading the file: {str(e)}")
return "" # Return empty string if error occurs
class ExampleVisualizationArgs(BaseModel):
query: str = Field(description="The user's query about plotting.")
example_visualization_tool = StructuredTool.from_function(
func=get_example_of_visualizations,
name="get_example_of_visualizations",
description="Retrieves example visualization code related to the user's query.",
args_schema=ExampleVisualizationArgs
)
########################################
# 1) DEFINE THE TOOL FOR LISTING FILES #
########################################
class ListPlottingDataFilesArgs(BaseModel):
# No arguments needed here if it just lists everything
dummy: str = Field(default="", description="(No arguments needed)")
def list_plotting_data_files(dummy: str = "") -> str:
"""
Lists all files and subdirectories under data/plotting_data.
Returns a single string containing each path on a new line.
"""
base_dir = os.path.join("data", "plotting_data")
all_paths = []
for root, dirs, files in os.walk(base_dir):
# Optionally skip hidden dirs/files, etc.
for filename in files:
rel_path = os.path.relpath(os.path.join(root, filename), start=base_dir)
all_paths.append(rel_path)
if not all_paths:
return "No files found in data/plotting_data."
return "Files under data/plotting_data:\n" + "\n".join(all_paths)
list_plotting_data_files_tool = StructuredTool.from_function(
func=list_plotting_data_files,
name="list_plotting_data_files",
description="Lists all files under data/plotting_data directory (including subfolders).",
args_schema=ListPlottingDataFilesArgs
)
def create_visualization_agent(user_query, datasets_info):
datasets_text = "" # Initialize datasets_text
dataset_variables = []
datasets = {}
for i, info in enumerate(datasets_info):
var_name = f"dataset_{i + 1}"
datasets[var_name] = info['dataset']
dataset_variables.append(var_name)
# Build datasets_text
datasets_text += (
f"Dataset {i + 1}:\n"
f"Variable Name: {var_name}\n"
f"Name: {info['name']}\n"
f"Description: {info['description']}\n"
f"Head of DataFrame (use it only as an example):\n"
f"{info['df_head']}\n\n"
)
# Generate the system prompt using datasets_text
prompt = Prompts.generate_visualization_agent_system_prompt(user_query, datasets_text, dataset_variables)
llm = ChatOpenAI(api_key=API_KEY, model_name=st.session_state.model_name)
repl_tool = CustomPythonREPLTool(datasets=datasets)
tools_vis = [
repl_tool,
reflect_tool,
install_package_tool,
example_visualization_tool,
list_plotting_data_files_tool
]
agent_visualization = create_openai_tools_agent(
llm,
tools=tools_vis,
prompt=ChatPromptTemplate.from_messages(
[
("system", prompt),
MessagesPlaceholder(variable_name="messages"),
MessagesPlaceholder(variable_name="agent_scratchpad")
]
)
)
return AgentExecutor(
agent=agent_visualization,
tools=tools_vis,
verbose=True,
handle_parsing_errors=True,
return_intermediate_steps=True
)
def create_hard_coded_visualization_agent(user_query, datasets_info):
import streamlit as st
model_name = st.session_state.get("model_name", "gpt-3.5-turbo")
if model_name == "o3-mini":
llm = ChatOpenAI(api_key=API_KEY, model_name=model_name)
else:
llm = ChatOpenAI(api_key=API_KEY, model_name=model_name)
# Prepare datasets
datasets = {}
datasets_text = ""
dataset_variables = []
for i, info in enumerate(datasets_info):
var_name = f"dataset_{i + 1}"
datasets[var_name] = info['dataset']
dataset_variables.append(var_name)
datasets_text += (
f"Dataset {i + 1}:\n"
f"Variable Name: {var_name}\n"
f"Name: {info['name']}\n"
f"Description: {info['description']}\n"
f"Head of DataFrame (select appropriate attributes based on this):\n"
f"{info['df_head']}\n\n"
)
# Generate the system prompt
system_prompt = Prompts.generate_system_prompt_hard_coded_visualization(user_query, datasets_text, dataset_variables)
def plot_master_track_map_tool(dataset_var, main_title, lat_col, lon_col, date_col):
dataset_df = datasets.get(dataset_var)
if dataset_df is None:
return {"result": f"Dataset '{dataset_var}' not found."}
return plot_master_track_map(main_title=main_title, lat_col=lat_col, lon_col=lon_col, date_col=date_col, dataset_df=dataset_df)
# Define visualization tools
visualization_functions = [
StructuredTool.from_function(
func=plot_master_track_map_tool,
name="plot_master_track_map_tool",
description="Plot the master track map using the specified dataset.",
args_schema=PlotMasterTrackMapArgs
)
]
# Create the agent with tools and prompt
agent = create_openai_tools_agent(
llm,
tools=visualization_functions,
prompt=ChatPromptTemplate.from_messages(
[
("system", system_prompt),
MessagesPlaceholder(variable_name="messages"),
MessagesPlaceholder(variable_name="agent_scratchpad")
]
)
)
return AgentExecutor(agent=agent, tools=visualization_functions)
# Create Oceanographer Agent
def create_oceanographer_agent(user_query, datasets_info):
import streamlit as st
model_name = st.session_state.get("model_name", "gpt-3.5-turbo")
if model_name == "o3-mini":
llm = ChatOpenAI(api_key=API_KEY, model_name=model_name)
else:
llm = ChatOpenAI(api_key=API_KEY, model_name=model_name)
# Prepare datasets
datasets = {}
datasets_text = ""
dataset_variables = []
for i, info in enumerate(datasets_info):
var_name = f"dataset_{i + 1}"
datasets[var_name] = info['dataset']
dataset_variables.append(var_name)
datasets_text += (
f"Dataset {i + 1}:\n"
f"Variable Name: {var_name}\n"
f"Name: {info['name']}\n"
f"Description: {info['description']}\n"
f"Head of DataFrame (select appropriate attributes based on this):\n"
f"{info['df_head']}\n\n"
)
# Generate the system prompt
system_prompt = Prompts.generate_system_prompt_oceanographer(user_query, datasets_text, dataset_variables)
def plot_ts_diagram_tool(dataset_var, main_title, temperature_col, salinity_col):
dataset_df = datasets.get(dataset_var)
if dataset_df is None:
return {"result": f"Dataset '{dataset_var}' not found."}
return plot_ts_diagram(main_title=main_title, temperature_col=temperature_col, salinity_col=salinity_col, dataset_df=dataset_df)
# Define oceanography tools
oceanography_functions = [
StructuredTool.from_function(
func=plot_ts_diagram_tool,
name="plot_ts_diagram_tool",
description="Plot TS diagram using the specified dataset.",
args_schema=TSPlotToolArgs
)
]
# Create the agent with tools and prompt
agent = create_openai_tools_agent(
llm,
tools=oceanography_functions,
prompt=ChatPromptTemplate.from_messages(
[
("system", system_prompt),
MessagesPlaceholder(variable_name="messages"),
MessagesPlaceholder(variable_name="agent_scratchpad")
]
)
)
return AgentExecutor(agent=agent, tools=oceanography_functions)
def initialize_agents(user_query, datasets_info):
if datasets_info:
# Create agents
visualization_agent = create_visualization_agent(
user_query=user_query,
datasets_info=datasets_info
)
dataframe_agent = create_pandas_agent(
user_query=user_query,
datasets_info=datasets_info
)
hard_coded_visualization_agent = create_hard_coded_visualization_agent(
user_query=user_query,
datasets_info=datasets_info
)
oceanographer_agent = create_oceanographer_agent(
user_query=user_query,
datasets_info=datasets_info
)
return visualization_agent, dataframe_agent, hard_coded_visualization_agent, oceanographer_agent
else:
st.warning("No datasets loaded. Please load datasets first.")
return None, None, None, None
def agent_node(state, agent, name):
import streamlit as st # Ensure Streamlit is imported
logging.debug(f"Entering agent_node for {name}")
if 'agent_scratchpad' not in state or not isinstance(state['agent_scratchpad'], list):
state['agent_scratchpad'] = []
user_messages = [msg for msg in state["messages"] if isinstance(msg, HumanMessage)]
if user_messages:
last_user_message = user_messages[-1].content
state['input'] = last_user_message
else:
state['input'] = state.get('input', '')
if 'plot_images' not in state or not isinstance(state['plot_images'], list):
state['plot_images'] = []
# Invoke the agent
result = agent.invoke(state)
last_message_content = result.get("output", "")
intermediate_steps = result.get("intermediate_steps", [])
returned_plot_images = result.get("plot_images", []) # Gather newly returned images
# Store intermediate steps
if 'intermediate_steps' not in st.session_state:
st.session_state['intermediate_steps'] = []
st.session_state['intermediate_steps'].extend(intermediate_steps)
from src.utils import log_history_event
for step in intermediate_steps:
action = step[0]
observation = step[1]
tool_name = action.tool
tool_input = action.tool_input
log_history_event(
st.session_state,
"tool_usage",
{
"agent_name": name,
"tool_name": tool_name,
"tool_input": tool_input,
"observation": observation
}
)
# If a ModuleNotFoundError was returned
if name == "VisualizationAgent":
if isinstance(last_message_content, dict):
if last_message_content.get("error") == "ModuleNotFoundError":
missing_module = last_message_content.get("missing_module")
logging.info(f"Detected missing module: {missing_module}")
install_result = install_package_tool.run({"package_name": missing_module})
logging.info(f"Install package result: {install_result}")
if "successfully" in install_result:
retry_result = agent.invoke(state)
last_message_content = retry_result.get("output", "")
else:
last_message_content = f"Failed to install the missing package '{missing_module}'. Please install it manually."
# Check if a new plot path was set in session_state
new_plot_path = st.session_state.get("new_plot_path")
logging.info(f"New plot path from session state: {new_plot_path}")
if new_plot_path:
if os.path.exists(new_plot_path):
state["plot_images"].append(new_plot_path)
st.session_state.new_plot_path = None
log_history_event(
st.session_state,
"plot_generated", # Use consistent event type
{
"plot_path": new_plot_path,
"agent_name": name,
"description": f"Plot generated by {name}"
}
)
if new_plot_path:
log_history_event(
st.session_state,
"plot_generated_final",
{"plot_path": new_plot_path}
)
# Combine the newly returned images with state images
all_plot_images = list(returned_plot_images) + state["plot_images"]
# Create a new AIMessage with additional info.
# Note: We add a "plot" field so that it appears in the final JSON.
ai_message = AIMessage(
content=last_message_content,
name=name,
additional_kwargs={
"plot_images": all_plot_images,
"plot": all_plot_images[0] if all_plot_images else None
}
)
state["messages"].append(ai_message)
# Trim messages if needed
state["messages"] = state["messages"][-7:]
if name == "VisualizationAgent":
state["visualization_agent_used"] = True
state["last_agent_message"] = last_message_content
return state
def supervisor_response(state):
import streamlit as st
from main import get_datasets_info_for_active_datasets # Adjust import as needed
model_name = st.session_state.get("model_name", "gpt-3.5-turbo")
if model_name == "o3-mini":
llm = ChatOpenAI(api_key=API_KEY, model_name=model_name)
else:
llm = ChatOpenAI(api_key=API_KEY, model_name=model_name)
# Build dataset context from the active (selected) datasets only.
active_datasets_info = get_datasets_info_for_active_datasets(st.session_state)
datasets_text = ""
if active_datasets_info:
for i, info in enumerate(active_datasets_info, 1):
datasets_text += (
f"Dataset {i}:\n"
f"Name: {info['name']}\n"
f"DOI: {info['doi']}\n"
f"Description: {info['description']}\n"
f"Parameters: {info.get('parameters', '')}\n\n"
)
else:
datasets_text = "No active dataset selected."
# Build the system prompt using the active dataset context.
system_message = (
"You are a supervisor capable of answering simple questions directly. "
"If the user's query is basic (e.g., about available analysis), "
"answer using the selected dataset context below:\n\n"
f"{datasets_text}\n\n"
"For complex queries, follow these agent guidelines:\n"
"- Use VisualizationAgent for general plotting\n"
"- Use HardCodedVisualizationAgent ONLY for track maps\n"
"- Use OceanographerAgent ONLY for TS diagrams\n"
"Format any code in markdown and keep responses concise."
)
# Build the complete conversation history.
# Here we include both human and assistant messages with labels.
full_history = "\n".join([
f"{msg.name}: {msg.content}" for msg in state["messages"] if hasattr(msg, "content") and hasattr(msg, "name")
])
prompt = f"{system_message}\n\nConversation history:\n{full_history}"
# Invoke the LLM with the full conversation context.
response = llm.invoke([HumanMessage(content=prompt)])
# Append the supervisor's answer to the state and mark the conversation as finished.
state["messages"].append(AIMessage(content=response.content, name="Supervisor"))
state["next"] = "FINISH"
return state
def create_supervisor_agent(user_query, datasets_info, memory):
members = ["VisualizationAgent", "DataFrameAgent", "HardCodedVisualizationAgent", "OceanographerAgent"]
# Prepare datasets_text and dataset_variables
datasets_text = ""
dataset_variables = []
datasets = {}
for i, info in enumerate(datasets_info):
var_name = f"df{i}" if i > 0 else "df"
datasets_text += (
f"Dataset {i + 1}:\n"
f"Variable Name: {var_name}\n"
f"Name: {info['name']}\n"
f"Description: {info['description']}\n"
f"Head of DataFrame (use it only as an example):\n"
f"{info['df_head']}\n\n"
)
dataset_variables.append(var_name)
datasets[var_name] = info['dataset']
system_prompt_supervisor = (
f"You are a supervisor tasked with managing a conversation between the following workers: {members}. "
f"Given the following user request: '{user_query}', determine and instruct the next worker to act. "
f"Each worker will perform a task and respond with their results and status. "
f"If the request involves plotting a master track, directly assign the task to the HardCodedVisualizationAgent. "
f"For TS diagram, assign the task to the OceanographerAgent. The other requests should be handled by the VisualizationAgent. It is extremely important to assign the correct task to the correct agent and use HardCodedVisualizationAgent and OceanographerAgent only for the described cases.\n"
f"If a meaningful response from the agent has been provided, end the process by returning 'FINISH' and not 'RESPOND' to avoid unnecessary loops.\n"
f"The dataset info is:\n{datasets_text}\n"
f"### Agents and Their Capabilities:\n"
"- **VisualizationAgent:** A major visualization tool to be called. Generates various plots using the dataset with tools like Python_REPL, reflect_on_image, install_package, and get_example_of_visualizations.\n"
"- **DataFrameAgent:** Performs data analysis and manipulation on the dataset using pandas.\n"
"- **HardCodedVisualizationAgent:** Only can plot master track map using predefined functions (call only if you are 100% sure that you need a master track map from an expedition; otherwise, call VisualizationAgent).\n"
"- **OceanographerAgent:** Only can plot TS diagrams (call only if you are 100% sure that you need to create a TS diagram; otherwise, call VisualizationAgent).\n\n"
f"### Available Tools:\n"
f"- **Python_REPL:** Executes Python code for data analysis and visualization.\n"
f"- **reflect_on_image:** Provides feedback on generated images to improve their quality.\n"
f"- **install_package:** Installs necessary Python packages when encountering missing modules.\n"
f"- **get_example_of_visualizations:** Retrieves example visualization code related to user queries.\n"
f"\n"
f"The datasets are accessible via variables: {', '.join(dataset_variables)}.\n"
)
# Define the function for routing the next task
function_def = {
"name": "route",
"description": "Select the next role.",
"parameters": {
"title": "routeSchema",
"type": "object",
"properties": {
"next": {
"title": "Next",
"anyOf": [
{"enum": ["FINISH", "RESPOND"] + members},
],
}
},
"required": ["next"],
},
}
# Create the supervisor chain
prompt_supervisor = ChatPromptTemplate.from_messages(
[
("system", system_prompt_supervisor),
MessagesPlaceholder(variable_name="messages"),
MessagesPlaceholder(variable_name="agent_scratchpad"),
("system",
f"Given the conversation above, decide who should act next. Options are: ['FINISH', 'RESPOND'] + {members}.\n"
"Select 'FINISH' if the last agent has provided a meaningful and complete response to the user's query.\n"
"Select 'RESPOND' if you need to provide additional information or clarification to the user.\n"
"Otherwise, select the next agent to act.\n"
f"The last agent message was: {{last_agent_message}}")
]
).partial(options=str(["FINISH", "RESPOND"] + members), members=", ".join(members))
llm_supervisor = ChatOpenAI(api_key=API_KEY, model_name=st.session_state.model_name)
supervisor_chain = (
{
"messages": lambda x: x["messages"],
"agent_scratchpad": lambda x: x["agent_scratchpad"],
"last_agent_message": lambda x: x.get("last_agent_message", ""),
}
| prompt_supervisor
| llm_supervisor.bind_functions(functions=[function_def], function_call="route")
| JsonOutputFunctionsParser()
)
# Define the AgentState type
class AgentState(TypedDict):
messages: Sequence[BaseMessage]
next: str
agent_scratchpad: Sequence[BaseMessage]
user_query: str
last_agent_message: str
plot_images: List[str]
model_name: str
# Create the workflow graph
workflow = StateGraph(AgentState)
visualization_agent, dataframe_agent, hard_coded_visualization_agent, oceanographer_agent = initialize_agents(
user_query, datasets_info
)
# Add agents to the workflow if they are successfully initialized
if visualization_agent and dataframe_agent and hard_coded_visualization_agent and oceanographer_agent:
workflow.add_node("VisualizationAgent",
functools.partial(agent_node, agent=visualization_agent, name="VisualizationAgent"))
workflow.add_node("DataFrameAgent", functools.partial(agent_node, agent=dataframe_agent, name="DataFrameAgent"))
workflow.add_node("HardCodedVisualizationAgent",
functools.partial(agent_node, agent=hard_coded_visualization_agent,
name="HardCodedVisualizationAgent"))
workflow.add_node("OceanographerAgent",
functools.partial(agent_node, agent=oceanographer_agent, name="OceanographerAgent"))
workflow.add_node("supervisor", supervisor_chain)
workflow.add_node("supervisor_response", supervisor_response)
# Connect agents to the supervisor
for member in members:
workflow.add_edge(member, "supervisor")
# Define the conditional map for routing
conditional_map = {k: k for k in members}
conditional_map["FINISH"] = END
conditional_map["RESPOND"] = "supervisor_response"
workflow.add_conditional_edges("supervisor", lambda x: x["next"], conditional_map)
workflow.set_entry_point("supervisor")
#memory = MemorySaver()
# Compile the workflow into a graph
graph = workflow.compile(checkpointer=memory)
return graph
else:
return None |