Neurosymbolic QA with WolframAlpha API

docs/sphinx_doc/en/source/tutorial/204-service.md

@@ -49,7 +49,10 @@ The following table outlines the various Service functions by type. These functi
 |                             | `openai_image_to_text` | Convert text to image using OpenAI API
 |                             | `openai_text_to_audio` | Convert text to audio using OpenAI API
 |                             | `openai_audio_to_text` | Convert audio to text using OpenAI API
-
+| Reasoning                   | `query_wolfram_alpha_short_answers`  | Query the Wolfram Alpha Short Answers API.                                                           |
+|                             | `query_wolfram_alpha_simple`         | Query the Wolfram Alpha Simple API.                                                                  |
+|                             | `query_wolfram_alpha_show_steps`     | Query the Wolfram Alpha Show Steps API.                                                              |
+|                             | `query_wolfram_alpha_llm`            | Query the Wolfram Alpha LLM API.                                                                     |
 
 | *More services coming soon* |                            | More service functions are in development and will be added to AgentScope to further enhance its capabilities. |
 

docs/sphinx_doc/zh_CN/source/tutorial/204-service.md

@@ -46,6 +46,10 @@
 |                             | `openai_image_to_text` | 使用 OpenAI API 根据图片生成文字。
 |                             | `openai_text_to_audio` | 使用 OpenAI API 根据文本生成音频。
 |                             | `openai_audio_to_text` | 使用OpenAI API将音频转换为文本。
+| 推理                         | query_wolfram_alpha_short_answers  | 查询 Wolfram Alpha Short Answers API。                |
+|                             | query_wolfram_alpha_simple         | 查询 Wolfram Alpha Simple API。                       |
+|                             | query_wolfram_alpha_show_steps     | 查询 Wolfram Alpha Show Steps API。                   |
+|                             | query_wolfram_alpha_llm            | 查询 Wolfram Alpha LLM API。                          |
 | *更多服务即将推出* |                       | 正在开发更多服务功能，并将添加到 AgentScope 以进一步增强其能力。  |
 
 关于详细的参数、预期输入格式、返回类型，请参阅[API文档](https://modelscope.github.io/agentscope/)。

examples/neurosymbolic_QA/README.md

@@ -0,0 +1,35 @@
+
+
+# Neurosymbolic Mathematical Reasoning Question Answering with AgentScope
+
+This example will show:
+- How to set up and use the ReActAgent from AgentScope for neurosymbolic reasoning.
+- How to integrate various Wolfram Alpha query functions into a neurosymbolic question-answering system, particularly for question involves mathetical reasoning.
+- How to use AgentScope Studio to visualize conversations between the agents.
+
+## Background
+LLMs excel at many tasks, but currently still fall short in rigorous reasoning, like doing complex mathematical calcuations.
+This script demonstrates a neurosymbolic question-answering system using AgentScope, which integrates various external APIs like Wolfram Alpha to assist LLM agents in solving problems that involve mathetical reasoning. The LLM agent first formulates the mathematical problems that need to be solved based on its interpretation of the problem statement, then it uses Wolfram Alpha to solve the mathematical problems, synthesizes the solution and finally returns a reply.
+
+Two LLM agents are set up in this demonstration: one with access to Wolfram Alpha APIs for solving mathematical problems, and another without such access (for comparison), aiming to solve a game theory problem on finding mixed Nash equilibria. The agent without access to the tools will often fail to give the correct solution.
+
+
+
+## Tested Models
+
+These models are tested in this example. For other models, some modifications may be needed.
+- `gpt-4o-mini` - With the given temperature setting. Note that with temperature tuning or more advanced models such as claude-3.5-sonnet, the agent is able to solve the given problem without relying on external tools. The example servers as a proof-of-concept for the scenario that, given a sufficiently different question for the given model capability, using external tools can increase the chance of successfully solving the problem,
+
+
+## Prerequisites
+
+To run this example successfully, ensure you meet the following requirements:
+- The latest AgentScope library installed.
+- Valid API keys for Wolfram Alpha and Anthropic API, which should be replaced in the placeholders within the script.
+- [Optional] Knowledge in game theory and understanding of Nash equilibria for interpreting the agent's reasoning trace.
+
+
+
+## Running the Example
+This example has been set to run with AgentScope Studio by default. After starting running the script, open `http://127.0.0.1:5000` with your browser (Chorme is preferred).
+For more information on AgentScope Studio see `http://doc.agentscope.io/en/tutorial/209-gui.html`

examples/neurosymbolic_QA/neurosymbolic_QA.py

@@ -0,0 +1,121 @@
+# -*- coding: utf-8 -*-
+"""
+This module implements a neurosymbolic question-answering system
+using AgentScope.
+
+It sets up two student agents: one with access to Wolfram Alpha APIs
+for solving mathematical problems, and another without such access.
+The agents are tasked with solving game theory problems,
+specifically finding mixed Nash equilibria.
+
+The module demonstrates the use of ReActAgent, ServiceToolkit, and various
+Wolfram Alpha query functions from AgentScope.
+"""
+from agentscope.service import ServiceToolkit
+from agentscope.message import Msg
+import agentscope
+from agentscope.agents import DialogAgent
+from agentscope.service import (
+    query_wolfram_alpha_short_answers,
+    query_wolfram_alpha_simple,
+    query_wolfram_alpha_show_steps,
+    query_wolfram_alpha_llm,
+)
+from agentscope.agents import ReActAgent
+
+
+agentscope.init(
+    # ...
+    project="xxx",
+    name="xxx",
+    studio_url="http://127.0.0.1:5000",  # The URL of AgentScope Studio
+)
+
+
+YOUR_MODEL_CONFIGURATION = [
+    {
+        "model_type": "openai_chat",
+        "config_name": "gpt",
+        "model_name": "gpt-4o-mini",
+        "api_key": "",
+        "generate_args": {
+            "temperature": 0.1,
+        },
+    },
+]
+
+
+# Initialize the ServiceToolkit and register the TripAdvisor API functions
+service_toolkit = ServiceToolkit()
+service_toolkit.add(
+    query_wolfram_alpha_short_answers,
+    api_key="",
+)  # Replace with your actual TripAdvisor API key
+service_toolkit.add(
+    query_wolfram_alpha_simple,
+    api_key="",
+)  # Replace with your actual TripAdvisor API key
+service_toolkit.add(
+    query_wolfram_alpha_show_steps,
+    api_key="",
+)  # Replace with your actual TripAdvisor API key
+service_toolkit.add(
+    query_wolfram_alpha_llm,
+    api_key="",
+)  # Replace with your actual TripAdvisor API key
+
+agentscope.init(model_configs=YOUR_MODEL_CONFIGURATION)
+
+student_agent1 = DialogAgent(
+    name="Student1",
+    sys_prompt="""You are a smart student. You are given problem to solve.""",
+    model_config_name="gpt",  # replace by your model config name
+)
+
+student_agent2 = ReActAgent(
+    name="Student2",
+    sys_prompt="""You are a smart student. You are given problem to solve.
+    Use the approprite wolfram alpha APIs to help you solve simultaneous
+    equations/calculations. Note that wolfram alpha apis can only help you
+    solve an explicitly given equation or calculation. Don't use them
+    if you are not solving equations or doing calculations.""",
+    model_config_name="gpt",
+    service_toolkit=service_toolkit,
+    verbose=True,  # set verbose to True to show the reasoning trace
+)
+
+
+x1 = Msg(
+    "system",
+    """Solve the problem: For the following games,
+    compute all mixed Nash equilibria. In each table cell,
+    the number on the left is the payoff to the row player, and
+    the number on the right is the payoff to the column player.
+        |   | L   | R   |
+        |---|-----|-----|
+        | T | 7,1 | 2,2 |
+        | B | 0,5 | 3,1 |""",
+    role="user",
+)
+
+student_agent1.speak(x1)
+x1 = student_agent1(x1)
+
+
+x2 = Msg(
+    "system",
+    """Solve the problem: For the following games,
+    compute all mixed Nash equilibria. In each table cell,
+    the number on the left is the payoff to the row player, and
+    the number on the right is the payoff to the column player.
+        |   | L   | R   |
+        |---|-----|-----|
+        | T | 7,1 | 2,2 |
+        | B | 0,5 | 3,1 |
+    You need to formulate the relevant equations
+    and ask it to solve it for you.""",
+    role="user",
+)
+
+student_agent2.speak(x2)
+x2 = student_agent2(x2)

src/agentscope/service/__init__.py

@@ -26,6 +26,12 @@
     dblp_search_authors,
     dblp_search_venues,
 )
+from .reasoning.wolfram_alpha import (
+    query_wolfram_alpha_short_answers,
+    query_wolfram_alpha_simple,
+    query_wolfram_alpha_show_steps,
+    query_wolfram_alpha_llm,
+)
 from .multi_modality.dashscope_services import (
     dashscope_image_to_text,
     dashscope_text_to_image,
@@ -101,6 +107,10 @@ def get_help() -> None:
     "openai_image_to_text",
     "openai_edit_image",
     "openai_create_image_variation",
+    "query_wolfram_alpha_short_answers",
+    "query_wolfram_alpha_simple",
+    "query_wolfram_alpha_show_steps",
+    "query_wolfram_alpha_llm",
     # to be deprecated
     "ServiceFactory",
 ]