Trabalho final de Reinforcement Learning

Em models temos uma estrutura baseada em:

1. Coders:

   • read_data(data, prompt) -> (cleaned_data, used_code)
   • improvement(data, improve_prompt, used_prompt, used_code) -> (cleaned_data, result_code)

2. Reviwer:

   • evaluation(cleaned_data, used_code, score) -> new_prompt
   • make_report(report_prompt, prompt_1, score_1, prompt_2, score_2) -> report

3. Judge:

   • evaluation(report) -> (score)

4. Enviorment:

   • gen_codes(data, prompt_1, prompt_2) -> (cleaned_data_1, code_1, cleaned_data_2, code_2)

   • score(cleaned_data) -> score

   • make_evaluation(cleaned_data, used_code, score) -> new_prompt

   • train_code(initial_prompt_c1, initial_prompt_c2, initial_prompt_r, n_iterations) -> (report, score_loss)

   • train( initial_prompt_c1, initial_prompt_c2, initial_prompt_r, report_prompt, n_iterations_code, n_iterations_report) -> (report, score_report, score_loss_coders)