I have created a Pandas Dataframe Agent with the LLM being a HuggingFace pipeline. I want the agent to use a multi-input StructuredTool, but I always get a ValidationError. I tried to follow the multi-input tool guide here and pass the Action Input as a JSON string but that doesn't work either.
The tool is defined like so:
from langchain.tools import tool
@tool()
def plot_graph(x: str, y: str):
r"""Plots a scatterplot of the x and y column values of the Dataframe. Provide the x and y arguments as a JSON object. Example: {{"x": "Amount", "y": "Date"}}"""
print(x, y) # placeholder
tools = [plot_graph]
REPL output:
[StructuredTool(name='plot_graph', description='plot_graph(x: str, y: str) - Plots a scatterplot of the x and y column values of the Dataframe. Provide the x and y arguments as a JSON object. Example: {{"x": "Amount", "y": "Date"}}', args_schema=<class 'pydantic.v1.main.plot_graphSchemaSchema'>, return_direct=False, verbose=False, callbacks=None, callback_manager=None, tags=None, metadata=None, handle_tool_error=False, func=<function plot_graph at 0x0000018294D30680>, coroutine=None)]
The agent is defined and run like this:
agent = create_pandas_dataframe_agent(hf, df, verbose=True, extra_tools=tools, agent=AgentType.STRUCTURED_CHAT_ZERO_SHOT_REACT_DESCRIPTION)
query = "Plot a graph of Credit Amount vs Date."
agent.run(query)
Note: I tried running the above without agent=AgentType.STRUCTURED_CHAT_ZERO_SHOT_REACT_DESCRIPTION as well.
Which leads the model to generate this:
Thought: I should now plot a graph of Credit Amount vs Date.
Action: plot_graph
Action Input: {"x": "Date", "y": "Credit Amount"}
But then results in this error:
---------------------------------------------------------------------------
ValidationError Traceback (most recent call last)
Cell In[34], line 6
5 query = "Plot a graph of Credit Amount vs Date."
----> 6 agent.run(query)
File ~\Documents\Python\langchain\.venv\Lib\site-packages\langchain\chains\base.py:507, in Chain.run(self, callbacks, tags, metadata, *args, **kwargs)
505 if len(args) != 1:
506 raise ValueError("`run` supports only one positional argument.")
--> 507 return self(args[0], callbacks=callbacks, tags=tags, metadata=metadata)[
508 _output_key
509 ]
511 if kwargs and not args:
512 return self(kwargs, callbacks=callbacks, tags=tags, metadata=metadata)[
513 _output_key
514 ]
File ~\Documents\Python\langchain\.venv\Lib\site-packages\langchain\chains\base.py:312, in Chain.__call__(self, inputs, return_only_outputs, callbacks, tags, metadata, run_name, include_run_info)
310 except BaseException as e:
311 run_manager.on_chain_error(e)
--> 312 raise e
313 run_manager.on_chain_end(outputs)
314 final_outputs: Dict[str, Any] = self.prep_outputs(
315 inputs, outputs, return_only_outputs
316 )
File ~\Documents\Python\langchain\.venv\Lib\site-packages\langchain\chains\base.py:306, in Chain.__call__(self, inputs, return_only_outputs, callbacks, tags, metadata, run_name, include_run_info)
299 run_manager = callback_manager.on_chain_start(
300 dumpd(self),
301 inputs,
302 name=run_name,
303 )
304 try:
305 outputs = (
--> 306 self._call(inputs, run_manager=run_manager)
307 if new_arg_supported
308 else self._call(inputs)
309 )
310 except BaseException as e:
311 run_manager.on_chain_error(e)
File ~\Documents\Python\langchain\.venv\Lib\site-packages\langchain\agents\agent.py:1126, in AgentExecutor._call(self, inputs, run_manager)
1124 # We now enter the agent loop (until it returns something).
1125 while self._should_continue(iterations, time_elapsed):
-> 1126 next_step_output = self._take_next_step(
1127 name_to_tool_map,
1128 color_mapping,
1129 inputs,
1130 intermediate_steps,
1131 run_manager=run_manager,
1132 )
1133 if isinstance(next_step_output, AgentFinish):
1134 return self._return(
1135 next_step_output, intermediate_steps, run_manager=run_manager
1136 )
File ~\Documents\Python\langchain\.venv\Lib\site-packages\langchain\agents\agent.py:981, in AgentExecutor._take_next_step(self, name_to_tool_map, color_mapping, inputs, intermediate_steps, run_manager)
979 tool_run_kwargs["llm_prefix"] = ""
980 # We then call the tool on the tool input to get an observation
--> 981 observation = tool.run(
982 agent_action.tool_input,
983 verbose=self.verbose,
984 color=color,
985 callbacks=run_manager.get_child() if run_manager else None,
986 **tool_run_kwargs,
987 )
988 else:
989 tool_run_kwargs = self.agent.tool_run_logging_kwargs()
File ~\Documents\Python\langchain\.venv\Lib\site-packages\langchain\tools\base.py:314, in BaseTool.run(self, tool_input, verbose, start_color, color, callbacks, tags, metadata, run_name, **kwargs)
300 def run(
301 self,
302 tool_input: Union[str, Dict],
(...)
311 **kwargs: Any,
312 ) -> Any:
313 """Run the tool."""
--> 314 parsed_input = self._parse_input(tool_input)
315 if not self.verbose and verbose is not None:
316 verbose_ = verbose
File ~\Documents\Python\langchain\.venv\Lib\site-packages\langchain\tools\base.py:245, in BaseTool._parse_input(self, tool_input)
243 if input_args is not None:
244 key_ = next(iter(input_args.__fields__.keys()))
--> 245 input_args.validate({key_: tool_input})
246 return tool_input
247 else:
File ~\Documents\Python\langchain\.venv\Lib\site-packages\pydantic\v1\main.py:711, in BaseModel.validate(cls, value)
708 value = cls._enforce_dict_if_root(value)
710 if isinstance(value, dict):
--> 711 return cls(**value)
712 elif cls.__config__.orm_mode:
713 return cls.from_orm(value)
File ~\Documents\Python\langchain\.venv\Lib\site-packages\pydantic\v1\main.py:341, in BaseModel.__init__(__pydantic_self__, **data)
339 values, fields_set, validation_error = validate_model(__pydantic_self__.__class__, data)
340 if validation_error:
--> 341 raise validation_error
342 try:
343 object_setattr(__pydantic_self__, '__dict__', values)
ValidationError: 1 validation error for plot_graphSchemaSchema
y
field required (type=value_error.missing)
How do I make this work?