samuelae / associator Goto Github PK

The \texttt{ar_cross_targets()} produces cross-tables between variables in the \texttt{targets} and \texttt{participants} tables while respecting the retrieval counts from \texttt{responses}. This can be used, for instance, to evaluate the proportion of responses from each target cluster for different individual difference factors. The function takes the \texttt{associatoR} object as input and returns a table of class \texttt{tibble} containing grouped statistics. Per default, the function produces frequencies, but the user can request group-normalized counts using the \texttt{normalize} argument.

ar_cross_targets(ar_obj, 
                 participant_vars = c(gender, education), 
                 target_var = cluster, 
                 normalize = TRUE)

when running the examples with the new and updated data set, the ar_compare_embeddings produces an error Error in svd(A, nu = nu, nv = nv) : infinite or missing values in 'x', but only if both participant vars are compared simultaneously

ar_obj = ar_import(intelligence,
                   participant = participant_id,
                   cue = cue,
                   response = response,
                   participant_vars = c(gender, education),
                   response_vars = c(response_position, response_level)) %>%
  ar_normalize(case = "most_frequent", punct = "end", whitespace = "squish", process_cues = TRUE) %>% 
  ar_set_targets(targets = "cues") %>%
  ar_embed_targets()
ar_compare_embeddings(ar_obj, c(gender, education))

the help file/documentation for ar_compare() needs a clear description of counting [what will be counted exactly] and distinguish clearly between summarizing and counting

remove them

when ar_cluster_targets() is run with an associatoR object containing a clustering solution already, multiple cluster variables (with suffixes) are joined together

• currently, 1, 2, 3, NA
• shall be replaced by obviously not numeric variable: cluster_1, cluster_2, cluster_3
• NA shall be replaced by something more informative: no_cluster

add exact prompt and system prompts, as well as GPT-4 version

remove the generic plot function for the associatoR class object, since there are now dedicated plotting functions (ar_plot_embedding() and ar_plot_wordcloud())

The \texttt{ar_compare_targets} function is the analog to \texttt{ar_cross_targets()} for continuous target variables. It can produce summaries of other variables, such as the targets' psycholinguistics properties. The function takes the \texttt{associatoR} object as input, as well as the participant and target variables of interest and, for the \texttt{fun} argument, a summarizing function. Per default, the function calculates the arithmetic mean. The function returns a table of class \texttt{tibble} containing grouped statistics

ar_compare_targets(ar_obj, 
                   participant_vars = c(gender, education), 
                   target_var = valence, 
                   fun = median)

to reproduce:

ar_obj <- ar_import(intelligence, 
                    participant = participant_id, 
                    cue = cue, 
                    response = response, 
                    participant_vars = c(gender, education), 
                    response_vars = c(response_position, response_level)) %>% 
  ar_normalize(case = "lower", punct = "all", whitespace = "squish") %>% 
  ar_set_targets("cues") %>% 
  ar_count_targets()
ar_correlate_targets(ar_obj, participant_vars = c("education", "gender"))

either ar_response_position(), or something more general

suggested by Dirk in 5. Characterizing targets

see manuscript

This reproduces the Error:

> ar_obj <- ar_import(intelligence, 
+                     participant = participant_id, 
+                     cue = cue, 
+                     response = response_clean, 
+                     participant_vars = c(gender, education), 
+                     response_vars = c(response_position, response_level))
Error in `dplyr::rename()`:
! Names must be unique.
✖ These names are duplicated:
  * "response" at locations 5 and 6.
Run `rlang::last_trace()` to see where the error occurred.
> rlang::last_trace()
<error/vctrs_error_names_must_be_unique>
Error in `dplyr::rename()`:
! Names must be unique.
✖ These names are duplicated:
  * "response" at locations 5 and 6.
---
Backtrace:
    ▆
 1. ├─associatoR::ar_import(...)
 2. │ └─data %>% dplyr::rename(id = !!participant) %>% ...
 3. ├─dplyr::rename(., response = !!response)
 4. └─dplyr:::rename.data.frame(., response = !!response)
Run rlang::last_trace(drop = FALSE) to see 12 hidden frames.

The \texttt{ar_compare_embeddings()} function evaluates the representational similarity of \texttt{target_embeddings} for each of one or more variables in the \texttt{participants} table. The group-specific \texttt{target_embeddings} are generated in the background and used to calculate the representational similarity based on either the correlation of the lower-triangle similarity matrices or the average correlation of row-similarities. This can be specified via the \texttt{use} argument. The function permits passing arguments to \texttt{ar_embed()} in the background via the ellipsis argument.

ar_compare_embeddings(ar_obj, 
                      participant_vars = c(gender, education), 
                      use = "triangle")

Add simple interface to correct spelling of words manually, or partially manually, e.g. check against dictionary, then provide interface to manually correct

make NULL or "original" or similar an option on all arguments