# Updated test for tabscreen_gpt.default functions test_that(".gpt_engine return errors correctly", { tools_choice_name <- list( type = "function", "function" = list( name = "inclusion_decision_simple" ) ) # Specifying unknown model body <- list( model = "gpt-7", messages = list(list( role = "user", content = question )), tools = tools_simple, tool_choice = tools_choice_name, top_p = 1 ) res <- .gpt_engine( body = body, RPM = 10000, timeinf = T, tokeninf = T, key = get_api_key(), max_t = 4, max_s = 10, is_trans = gpt_is_transient, back = NULL, aft = NULL ) expect_equal(ncol(res), 7) expect_true(stringr::str_detect(res$decision_gpt, "404")) # Specifying wrong api key body <- list( model = "gpt-4o", messages = list(list( role = "user", content = question )), tools = tools_simple, tool_choice = tools_choice_name, top_p = 1 ) res <- .gpt_engine( body = body, RPM = 10000, timeinf = T, tokeninf = T, key = 1234, max_t = 4, max_s = 10, is_trans = gpt_is_transient, back = NULL, aft = NULL ) expect_equal(ncol(res), 7) expect_true(stringr::str_detect(res$decision_gpt, "401")) # Specifying ineligible role body <- list( model = "gpt-4o", messages = list(list( role = "use", content = question )), tools = tools_simple, tool_choice = tools_choice_name, top_p = 1 ) res <- .gpt_engine( body = body, RPM = 10000, timeinf = T, tokeninf = T, key = get_api_key(), max_t = 4, max_s = 10, is_trans = gpt_is_transient, back = NULL, aft = NULL ) expect_equal(ncol(res), 7) expect_true(stringr::str_detect(res$decision_gpt, "400")) # Specifying ineligible role body <- list( model = "gpt-4o", messages = list(list( role = "use", content = question )), tools = tools_simple, tool_choice = tools_choice_name, top_p = 1 ) # Transient is correctly add to function .gpt_engine( body = body, RPM = 10000, timeinf = T, tokeninf = T, key = get_api_key(), max_t = 4, max_s = 10, is_trans = gpt_is_transient(), back = NULL, aft = NULL ) |> expect_error() }) test_that(".rep_gpt_engine controls errrors correctly", { iterations <- 3 # Ineligible role res <- .rep_gpt_engine( question = question, model_gpt = "gpt-4o-mini", topp = 1, role_gpt = "use", tool = tools_simple, t_choice = "inclusion_decision_simple", iterations = iterations, req_per_min = 10000, seeds = NULL, time_inf = T, token_inf = T, apikey = get_api_key(), maxt = 4, maxs = 10, istrans = gpt_is_transient, ba = NULL, af = NULL ) expect_equal(ncol(res), 8) expect_equal(nrow(res), iterations) expect_equal(max(res$n), iterations) expect_true(all(stringr::str_detect(res$decision_gpt, "400"))) # Ineligible api key res <- .rep_gpt_engine( question = question, model_gpt = "gpt-4o-mini", topp = 1, role_gpt = "user", tool = tools_simple, t_choice = "inclusion_decision_simple", iterations = iterations, req_per_min = 10000, seeds = NULL, time_inf = T, token_inf = T, apikey = 1, maxt = 4, maxs = 10, istrans = gpt_is_transient, ba = NULL, af = NULL ) expect_equal(ncol(res), 8) expect_equal(nrow(res), iterations) expect_equal(max(res$n), iterations) expect_true(all(stringr::str_detect(res$decision_gpt, "401"))) # Ineligible model res <- .rep_gpt_engine( question = question, model_gpt = "gpt-4o-min", topp = 1, role_gpt = "user", tool = tools_simple, t_choice = "inclusion_decision_simple", iterations = iterations, req_per_min = 10000, seeds = NULL, time_inf = T, token_inf = T, apikey = get_api_key(), maxt = 4, maxs = 10, istrans = gpt_is_transient, ba = NULL, af = NULL ) expect_equal(ncol(res), 8) expect_equal(nrow(res), iterations) expect_equal(max(res$n), iterations) expect_true(all(stringr::str_detect(res$decision_gpt, "404"))) }) # Old tests (tabscreen_gpt.original) paths <- system.file("extdata", "test_prompts.rds", package = "AIscreenR") prompts <- readRDS(file = paths) #prompts <- # purrr::map_chr( # paths, ~ { # readtext::readtext(.x)$text |> # stringr::str_remove_all("\n") # } # ) prompt <- prompts[1] prompt2 <- prompts[2] models <- c("gpt-3.5-turbo-0613", "gpt-4") reps <- c(10, 1) rpm <- c(3500, 200) skip <- TRUE skip_github_action <- FALSE future::plan(future::multisession) test_that("tabscreen_gpt() works with single parameter values.",{ if(skip) skip() if(skip_github_action) skip_on_ci() skip_on_cran() expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, title = title, abstract = abstract, model = "gpt-3.5-turbo-0613", reps = 1, max_tries = 0 ) ) expect_true(is.numeric(test_obj$answer_data_all$decision_binary)) expect_equal(nrow(test_obj$answer_data_all), 1L) expect_equal(nrow(test_obj$answer_data_sum), 1L) expect_equal(nrow(test_obj$price_data), 1L) expect_length(test_obj$price_dollar, 1L) expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = c("gpt-3.5-turbo-0613", "gpt-4"), reps = 1 ) ) expect_equal(nrow(test_obj$answer_data_all), 2L) expect_equal(nrow(test_obj$answer_data_sum), 2L) expect_equal(nrow(test_obj$price_data), 2L) expect_length(test_obj$price_dollar, 1L) expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = c(prompt, prompt2), studyid = studyid, title = title, abstract = abstract, model = c("gpt-3.5-turbo-0613"), reps = 1 ) ) expect_equal(nrow(test_obj$answer_data_all), 2L) expect_equal(nrow(test_obj$answer_data_sum), 2L) expect_equal(nrow(test_obj$price_data), 1L) expect_length(test_obj$price_dollar, 1L) }) test_that("tabscreen_gpt() works with multiple parameter values.",{ if(skip) skip() if(skip_github_action) skip_on_ci() skip_on_cran() expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = c("gpt-3.5-turbo-0613", "gpt-4"), reps = c(2, 1) ) ) expect_equal(nrow(test_obj$answer_data_all), 3L) expect_equal(nrow(test_obj$answer_data_sum), 2L) expect_equal(nrow(test_obj$price_data), 2L) expect_length(test_obj$price_dollar, 1L) expect_gt(nrow(test_obj$answer_data_all), nrow(test_obj$answer_data_sum)) expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1:2,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = c("gpt-3.5-turbo-0613", "gpt-4"), reps = c(2, 1) ) ) expect_equal(nrow(test_obj$answer_data_all), 6L) expect_equal(nrow(test_obj$answer_data_sum), 4L) expect_equal(nrow(test_obj$price_data), 2L) expect_length(test_obj$price_dollar, 1L) expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1:2,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = c("gpt-3.5-turbo-0613", "gpt-4"), reps = c(2, 1), top_p = c(0.2, 1) ) ) expect_equal(nrow(test_obj$answer_data_all), 12L) expect_equal(nrow(test_obj$answer_data_sum), 2*2*2) expect_equal(nrow(test_obj$price_data), 2L) expect_length(test_obj$price_dollar, 1L) }) test_that("tabscreen_gpt() don't return time and token info.", { if(skip) skip() if(skip_github_action) skip_on_ci() skip_on_cran() expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-3.5-turbo-0613", reps = 1, time_info = FALSE, token_info = FALSE ) ) time_token_not_include <- any(!stringr::str_detect(names(test_obj$answer_data_all), "time|token")) expect_true(time_token_not_include) expect_equal(length(test_obj), 2) expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-3.5-turbo-0613", reps = 2, api_key = 1234, time_info = FALSE, token_info = FALSE ) ) |> suppressMessages() expect_equal(length(test_obj), 3) expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-3.5-turbo-0613", reps = 1, time_info = TRUE, token_info = FALSE ) ) token_not_include <- any(!stringr::str_detect(names(test_obj$answer_data_all), "token")) expect_true(token_not_include) time_include <- any(stringr::str_detect(names(test_obj$answer_data_all), "time")) expect_true(time_include) expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-3.5-turbo-0613", reps = 1, time_info = FALSE, token_info = TRUE ) ) time_not_include <- any(!stringr::str_detect(names(test_obj$answer_data_all), "time")) expect_true(time_not_include) token_include <- any(stringr::str_detect(names(test_obj$answer_data_all), "token")) expect_true(token_include) }) test_that("tabscreen_gpt() works with detailed fucntions and ... .", { if(skip) skip() if(skip_github_action) skip_on_ci() skip_on_cran() expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1:2,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-3.5-turbo", reps = 1, functions = AIscreenR:::incl_function, function_call_name = list(name = "inclusion_decision") ) ) |> suppressMessages() expect_true(is.character(test_obj$answer_data_all$detailed_description)) expect_identical(nrow(test_obj$answer_data_all), nrow(test_obj$answer_data_sum)) complete_tokens <- mean(test_obj$answer_data_all$completion_tokens, na.rm = TRUE) expect_gt(complete_tokens, 13) # Working with max_tokens, i.e. ... calls expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-3.5-turbo", reps = 1, functions = AIscreenR:::incl_function, function_call_name = list(name = "inclusion_decision"), max_tokens = 40 ) ) |> suppressMessages() complete_tokens <- test_obj$answer_data_all$completion_tokens expect_lt(complete_tokens, 41) # Working with temperature, i.e. ... calls expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-3.5-turbo", reps = 3, functions = AIscreenR:::incl_function, function_call_name = list(name = "inclusion_decision"), progress = FALSE, temperature = 0.001 ) ) |> suppressMessages() n_distinct_detailed_answers <- test_obj$answer_data_all$detailed_description |> dplyr::n_distinct() expect_lt(n_distinct_detailed_answers, 3) }) test_that("Message behavior.", { if(skip) skip() if(skip_github_action) skip_on_ci() skip_on_cran() expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-4", reps = 1, messages = FALSE, progress = FALSE ) ) expect_no_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-3.5-turbo", reps = 1, messages = FALSE, progress = FALSE ) ) expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-4", reps = 2 ), "* Consider to reduce reps to 1 for gpt-4 models." ) expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-3.5-turbo", reps = 1, functions = AIscreenR:::incl_function, function_call_name = list(name = "inclusion_decision") ) ) |> suppressMessages() expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-4", reps = 1, functions = AIscreenR:::incl_function, function_call_name = list(name = "inclusion_decision") ), paste0( "\n* Be aware that getting detailed reponses from ChatGPT ", "will substantially increase the prize of the screening relative to the noted approximate prize." ) ) |> suppressMessages() expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[148,], prompt = prompt, studyid = studyid, title = title, abstract = abstract ), paste0( "* Consider removing references that has no abstract ", "since these can distort the accuracy of the screening" ) ) |> suppressMessages() }) test_that("tabscreen_gpt() expected errors.", { if(skip) skip() if(skip_github_action) skip_on_ci() skip_on_cran() expect_error( test_obj <- tabscreen_gpt( data = filges2015_dat[c(1:2),], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = models, reps = c(10, 1, 1) ) ) expect_error( test_obj <- tabscreen_gpt( data = filges2015_dat[c(1:2),], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = models, reps = c(10, 1), rpm = c(3500, 200, 200) ) ) expect_error( test_obj <- tabscreen_gpt( data = filges2015_dat[c(1:2),], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = c("gpt-3.5turbo-0613", "gpt-4"), reps = reps, rpm = rpm ) ) # With detaled description functions expect_error( test_obj <- tabscreen_gpt( data = filges2015_dat[c(1:2),], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = models, reps = c(10, 1, 1), functions = AIscreenR:::incl_function, function_call_name = list(name = "inclusion_decision") ) ) expect_error( test_obj <- tabscreen_gpt( data = filges2015_dat[c(1:2),], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = models, reps = c(10, 1), rpm = c(3500, 200, 200), functions = AIscreenR:::incl_function, function_call_name = list(name = "inclusion_decision") ) ) expect_error( test_obj <- tabscreen_gpt( data = filges2015_dat[c(1:2),], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = c("gpt-3.5turbo-0613", "gpt-4"), reps = reps, rpm = rpm, functions = AIscreenR:::incl_function, function_call_name = list(name = "inclusion_decision") ) ) expect_error( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, incl_cutoff_upper = 0.4, incl_cutoff_lower = 0.5 ) ) }) test_that("API error.",{ if(skip) skip() if(skip_github_action) skip_on_ci() skip_on_cran() expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, api_key = 1234, progress = FALSE ) ) |> suppressMessages() error_text <- unique(test_obj$answer_data_all$decision_gpt) expect_identical(error_text, "Error 401 Unauthorized [invalid api]") expect_no_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, api_key = 1234, messages = FALSE, progress = FALSE ) ) error_text <- unique(test_obj$answer_data_all$decision_gpt) expect_identical(error_text, "Error 401 Unauthorized [invalid api]") expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, api_key = 1234, progress = FALSE, functions = AIscreenR:::incl_function, function_call_name = list(name = "inclusion_decision") ) ) |> suppressMessages() error_text <- unique(test_obj$answer_data_all$decision_gpt) expect_identical(error_text, "Error 401 Unauthorized [invalid api]") expect_no_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, api_key = 1234, messages = FALSE, progress = FALSE, functions = AIscreenR:::incl_function, function_call_name = list(name = "inclusion_decision") ) ) error_text <- unique(test_obj$answer_data_all$decision_gpt) expect_identical(error_text, "Error 401 Unauthorized [invalid api]") }) future::plan(future::sequential) # Test parallel test_that("That paralell processing works.", { if(skip) skip() if(skip_github_action) skip_on_ci() skip_on_cran() future::plan(future::multisession, workers = future::availableCores()) expect_message( tm_par <- system.time( test_obj <- tabscreen_gpt( data = filges2015_dat[1:2,], prompt = prompt, title = title, abstract = abstract, model = "gpt-3.5-turbo-0613", reps = 10 ) ) ) future::plan(future::sequential) expect_lt(tm_par[["elapsed"]], sum(test_obj$answer_data_all$run_time)) gpt_answers <- test_obj$answer_data_all |> filter(!is.na(decision_gpt)) |> pull(decision_gpt) # Testing function call work gpt_answers |> stringr::str_detect("1|0|rror") |> unique() |> expect_true() }) test_that("max_tokens < 11 work",{ if(skip) skip() if(skip_github_action) skip_on_ci() skip_on_cran() expect_message( test_obj <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-3.5-turbo", reps = 1, max_tokens = 9 ) ) |> suppressMessages() expect_false(is.na(test_obj$answer_data_all$decision_gpt)) }) test_that("Print and class expectation for chatgpt object.", { if(skip) skip() if(skip_github_action) skip_on_ci() skip_on_cran() expect_message( random_name <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-3.5-turbo-0613", reps = 1, max_tries = 0 ) ) # class check expect_s3_class(random_name, "chatgpt") expect_s3_class(random_name, "list") expect_output(print(random_name), "random_name") expect_output(print.chatgpt(random_name), "random_name\\$answer_data_all") expect_output(print.chatgpt(random_name), "random_name\\$answer_data_sum") expect_output(print.chatgpt(random_name), "random_name\\$price_dollor") print_out1 <- paste0( "Find data with all answers by executing\\n random_name\\$answer_data_all\\n\\nFind ", "data with the result aggregated across multiple answers by executing\\n ", "random_name\\$answer_data_sum\\n\\nFind total price for the screening by executing\\n random_name\\$price_dollor" ) expect_output(print(random_name), print_out1) expect_message( random_name <- tabscreen_gpt( data = filges2015_dat[c(1:2),], prompt = c(prompt), studyid = studyid, # indicate the variable with the studyid in the data title = title, # indicate the variable with the titles in the data abstract = abstract, api_key = 1234 ) ) |> suppressMessages() expect_identical(names(random_name$answer_data_all), names(random_name$error_data)) expect_output(print(random_name), "random_name\\$answer_data_all") expect_output(print(random_name), "random_name\\$answer_data_sum") expect_output(print(random_name), "random_name\\$price_dollor") expect_output(print(random_name), "random_name\\$error_data") print_out2 <- paste0( "Find data with all answers by executing\\n random_name\\$answer_data_all\\n\\nFind ", "data with the result aggregated across multiple answers by executing\\n ", "random_name\\$answer_data_sum\\n\\nFind total price for the screening by executing\\n random_name\\$price_dollor" ) expect_output(print(random_name), print_out2) ############# expect_message( random_name <- tabscreen_gpt( data = filges2015_dat[1,], prompt = prompt, studyid = studyid, title = title, abstract = abstract, model = "gpt-3.5-turbo-0613", reps = 1, token_info = FALSE ) ) # class check expect_s3_class(random_name, "chatgpt") expect_s3_class(random_name, "list") expect_output(print(random_name), "random_name") expect_output(print.chatgpt(random_name), "random_name\\$answer_data_all") expect_output(print.chatgpt(random_name), "random_name\\$answer_data_sum") expect_false("price_dollor" %in% names(random_name)) expect_message( random_name <- tabscreen_gpt( data = filges2015_dat[c(1:2),], prompt = c(prompt), studyid = studyid, # indicate the variable with the studyid in the data title = title, # indicate the variable with the titles in the data abstract = abstract, api_key = 1234, token_info = FALSE ) ) |> suppressMessages() expect_output(print(random_name), "random_name\\$answer_data_all") expect_output(print(random_name), "random_name\\$answer_data_sum") expect_output(print(random_name), "random_name\\$error_data") expect_false("price_dollor" %in% names(random_name)) })