-
Notifications
You must be signed in to change notification settings - Fork 34
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Feat/prediction with rules and report #217
Feat/prediction with rules and report #217
Conversation
…rmation chunks and format additional_information string for prediction prompt
…x: Add function to check for well formatted output and values.
…craped text if it is too short: <300 characters.
…inal summarizer in the end of research tool.
…ating search queries.
@0xArdi Thanks for the instructions. I did all of this. It seems that the tool checks fail because the environment variables (API-keys) are not loaded in the workflow properly. Locally the tool checks succeed. I'm not sure how to fix this. |
@0xArdi Have you had this issue before? Do you know how to fix this? Does this maybe have something to do with the |
…h-rules-and-report
@jhehemann are you still experiencing this issue? Can you merge main and push? |
@jhehemann any update here? |
@0xArdi I merged main and pushed again. All checks pass except for the test. It seems that the api keys are not provided during workflow execution. If there is anything I can do, please let me know. |
We are in the process of migrating the project to a new location.
Thank you for your contributions! |
This pull request adds three tools to the mech agent. The prediction tool "prediction_with_rules_and_report" and the two supporting tools "infer_market_rules" and "research" that the prediction tool depends on.
RESEARCH:
This tool generates summaries of the most relevant content from the most relevant websites for answering the market question. The workflow is as follows:
INFER_MARKET_RULES:
It was observed that when gpt-3.5 is used to make predictions it struggles to conclude the correct outcome of a market question even if the event's date is mentioned in the search output. This tool infers status information and definitions that the question's phrasing implies and defines outcome rules based on the timing that is asked for in the question ("on", "by"). Few shot prompts are used here as it turned out that they manage to convey the subtle differences in phrasing the event, timeline, etc. best.
PREDICTION_WITH_RULES_AND_REPORT:
This tool uses the research tool and the infer_market_rules tool to get additional information. With this information a report is generated that analyses based on the research output as well as the status information and definitions from the infer_market_rules tool if and when the described event could happen. For this, the date is removed from the question to avoid any bias. A final prediction is made based on this report and the outcome rules from the infer_market_rules tool.