Skip to content

Latest commit

 

History

History
37 lines (36 loc) · 2.8 KB

WiP.md

File metadata and controls

37 lines (36 loc) · 2.8 KB
Raw

Work-in-Progress

  • Terminate the loop properly and avoid repeated run of loop.
  • Dataset Demo dataset of for data analysis from Kaggle. Mar. 3
  • ipynb report export @Mar 10
    • Export a ipynb type output from recorded analysis chat history! @Mar 10
    • support multi media outputs (table, image) in the captured output and export! @Mar 10
    • Export a ipynb type output from the system, record. Build up the ipynb in the process tool_chat_loop_2nb.
    • Support PDF and HTML formulated export of ipynb.
  • Presentation Nicer way to summarize results, code and figures into a nice looking report for human to read, instead of reading through the debugging history.
    • Jupyter notebook report
    • PDF and HTML report
    • Post processing and filtering of report.
    • Bug fix, if the loop in running in a terminal then the captured won't get the figure component! wont go to report!!
  • Folder structure Design file structure for storing analysis results.
    • reports sub-directory
    • notebook
    • export figure directory?
    • cleaned up code
  • Multi agent tool chat, one assigning tasks and objectives the other one execute it.
    • Code architecture for these multi agent system.
    • Roles
      • Analysis Supervisor. Breaking down the overall objective into smaller questions that could be analyzed by code, assign them to data analyst.
      • Data analyst. Taking the objective and tasks into account, using code to solve each one of them.
      • Visual analyst. Draw conclusions from reading the figures and code. @Mar.12th
      • Research Manager. Maintain a list of TODO for the analyst, hand out task one by one, when one analysis is finished check the result and write summary, assign a new one.
        • Format the objective in a structred list, as json etc. @Mar.12th
        • Then we can ask / answer them one by one? How to manage better?
      • Summary writer. Taking the output of data analyst and polish it into a report.
  • Vision interface, current agent cannot see the plot, so the analysis is not quite informative....
    • Demo for vision api drawing conclusions from the figures. @Mar.11.
    • Integrating vision agent with the main chat loop and integrating visual insight @Mar.12th.
    • Vision API is significantly slower than others.....
  • Overall CLI for default data analysis of a data set. Have a default workflow. Supervisor => Data analysis =>
    • Think about input structure. csvpath, column description, dataset description. overall objective
  • Add memory slot, let it remember the objective by reading it back.
  • Multi-file complex system analysis project.