Merge pull request #8 from edgararuiz/updates

Adds Ollama tests

.github/workflows/R-CMD-check.yaml

@@ -0,0 +1,52 @@
+# Workflow derived from https://github.com/r-lib/actions/tree/v2/examples
+# Need help debugging build failures? Start at https://github.com/r-lib/actions#where-to-find-help
+on:
+  push:
+    branches: main
+  pull_request:
+    branches: main
+
+name: R-CMD-check.yaml
+
+permissions: read-all
+
+jobs:
+  R-CMD-check:
+    runs-on: ${{ matrix.config.os }}
+
+    name: ${{ matrix.config.os }} (${{ matrix.config.r }})
+
+    strategy:
+      fail-fast: false
+      matrix:
+        config:
+          - {os: macos-latest,   r: 'release'}
+          - {os: windows-latest, r: 'release'}
+          - {os: ubuntu-latest,   r: 'devel', http-user-agent: 'release'}
+          - {os: ubuntu-latest,   r: 'release'}
+          - {os: ubuntu-latest,   r: 'oldrel-1'}
+
+    env:
+      GITHUB_PAT: ${{ secrets.GITHUB_TOKEN }}
+      R_KEEP_PKG_SOURCE: yes
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: r-lib/actions/setup-pandoc@v2
+
+      - uses: r-lib/actions/setup-r@v2
+        with:
+          r-version: ${{ matrix.config.r }}
+          http-user-agent: ${{ matrix.config.http-user-agent }}
+          use-public-rspm: true
+
+      - uses: r-lib/actions/setup-r-dependencies@v2
+        with:
+          extra-packages: any::rcmdcheck
+          needs: check
+
+      - uses: r-lib/actions/check-r-package@v2
+        with:
+          upload-snapshots: true
+          build_args: 'c("--no-manual","--compact-vignettes=gs+qpdf")'

NAMESPACE

@@ -32,6 +32,7 @@ import(glue)
 import(rlang)
 importFrom(dplyr,bind_cols)
 importFrom(dplyr,mutate)
+importFrom(dplyr,pull)
 importFrom(dplyr,tibble)
 importFrom(jsonlite,fromJSON)
 importFrom(ollamar,chat)

R/llm-classify.R

@@ -44,7 +44,7 @@ llm_classify.data.frame <- function(.data,
 llm_vec_classify <- function(x,
                              labels,
                              additional_prompt = "") {
-  llm_vec_prompt(
+  l_vec_prompt(
     x = x,
     prompt_label = "classify",
     additional_prompt = additional_prompt,

R/llm-custom.R

@@ -40,28 +40,9 @@ llm_custom.data.frame <- function(.data,
 #' @rdname llm_custom
 #' @export
 llm_vec_custom <- function(x, prompt = "", valid_resps = NULL) {
-  llm_use(.silent = TRUE, force = FALSE)
-  if (!inherits(prompt, "list")) {
-    p_split <- strsplit(prompt, "\\{\\{x\\}\\}")[[1]]
-    if (length(p_split) == 1 && p_split == prompt) {
-      content <- glue("{prompt}\n{{x}}")
-    } else {
-      content <- prompt
-    }
-    prompt <- list(list(role = "user", content = content))
-  }
-  resp <- m_backend_submit(defaults_get(), x, prompt)
-  if (!is.null(valid_resps)) {
-    errors <- !resp %in% valid_resps
-    resp[errors] <- NA
-    if (any(errors)) {
-      cli_alert_warning(
-        c(
-          "There were {sum(errors)} predictions with ",
-          "invalid output, they were coerced to NA"
-        )
-      )
-    }
-  }
-  resp
+  l_vec_prompt(
+    x = x,
+    prompt = prompt,
+    valid_resps = valid_resps
+  )
 }

R/llm-extract.R

@@ -31,8 +31,9 @@ llm_extract.data.frame <- function(.data,
                                    additional_prompt = "",
                                    pred_name = ".extract") {
   if (expand_cols && length(labels) > 1) {
+    text <- pull(.data, {{ col }})
     resp <- llm_vec_extract(
-      x = .data$col,
+      x = text,
       labels = labels,
       additional_prompt = additional_prompt
     )
@@ -76,7 +77,7 @@ llm_extract.data.frame <- function(.data,
 llm_vec_extract <- function(x,
                             labels = c(),
                             additional_prompt = "") {
-  resp <- llm_vec_prompt(
+  resp <- l_vec_prompt(
     x = x,
     prompt_label = "extract",
     labels = labels,

R/llm-sentiment.R

@@ -53,7 +53,7 @@ globalVariables("ai_analyze_sentiment")
 llm_vec_sentiment <- function(x,
                               options = c("positive", "negative", "neutral"),
                               additional_prompt = "") {
-  llm_vec_prompt(
+  l_vec_prompt(
     x = x,
     prompt_label = "sentiment",
     additional_prompt = additional_prompt,

R/llm-summarize.R

@@ -52,7 +52,7 @@ globalVariables("ai_summarize")
 llm_vec_summarize <- function(x,
                               max_words = 10,
                               additional_prompt = "") {
-  llm_vec_prompt(
+  l_vec_prompt(
     x = x,
     prompt_label = "summarize",
     additional_prompt = additional_prompt,

R/llm-translate.R

@@ -39,7 +39,7 @@ llm_vec_translate <- function(
     x,
     language,
     additional_prompt = "") {
-  llm_vec_prompt(
+  l_vec_prompt(
     x = x,
     prompt_label = "translate",
     additional_prompt = additional_prompt,

R/llm-use.R

@@ -7,8 +7,8 @@
 #' @param .silent Avoids console output
 #' @param model The name of model supported by the back-end provider
 #' @param ... Additional arguments that this function will pass down to the
-#' integrating function. In the case of Ollama, it will pass those argument to
-#' `ollamar::generate()`.
+#' integrating function. In the case of Ollama, it will pass those arguments to
+#' `ollamar::chat()`.
 #' @param force Flag that tell the function to reset all of the settings in the
 #' R session
 #'

R/m-backend-prompt.R

@@ -93,19 +93,55 @@ m_backend_prompt.mall_defaults <- function(backend, additional = "") {
   )
 }
 
-get_prompt <- function(label, ..., .additional = "") {
-  defaults <- m_backend_prompt(defaults_get(), additional = .additional)
-  fn <- defaults[[label]]
-  fn(...)
-}
-
-
-llm_vec_prompt <- function(x,
-                           prompt_label = "",
-                           additional_prompt = "",
-                           valid_resps = NULL,
-                           ...) {
-  llm_use(.silent = TRUE, force = FALSE)
-  prompt <- get_prompt(prompt_label, ..., .additional = additional_prompt)
-  llm_vec_custom(x, prompt, valid_resps = valid_resps)
+l_vec_prompt <- function(x,
+                         prompt_label = "",
+                         additional_prompt = "",
+                         valid_resps = NULL,
+                         prompt = NULL,
+                         ...) {
+  # Initializes session LLM
+  backend <- llm_use(.silent = TRUE, force = FALSE)
+  # If there is no 'prompt', then assumes that we're looking for a
+  # prompt label (sentiment, classify, etc) to set 'prompt'
+  if (is.null(prompt)) {
+    defaults <- m_backend_prompt(
+      backend = backend,
+      additional = additional_prompt
+    )
+    fn <- defaults[[prompt_label]]
+    prompt <- fn(...)
+  }
+  # If the prompt is a character, it will convert it to
+  # a list so it can be processed
+  if (!inherits(prompt, "list")) {
+    p_split <- strsplit(prompt, "\\{\\{x\\}\\}")[[1]]
+    if (length(p_split) == 1 && p_split == prompt) {
+      content <- glue("{prompt}\n{{x}}")
+    } else {
+      content <- prompt
+    }
+    prompt <- list(
+      list(role = "user", content = content)
+    )
+  }
+  # Submits final prompt to the LLM
+  resp <- m_backend_submit(
+    backend = backend,
+    x = x,
+    prompt = prompt
+  )
+  # Checks for invalid output and marks them as NA
+  if (!is.null(valid_resps)) {
+    errors <- !resp %in% valid_resps
+    resp[errors] <- NA
+    if (any(errors)) {
+      cli_alert_warning(
+        c(
+          "There were {sum(errors)} predictions with ",
+          "invalid output, they were coerced to NA"
+        )
+      )
+    }
+  }
+  resp
 }

R/m-backend-submit.R

@@ -32,21 +32,13 @@ m_backend_submit.mall_ollama <- function(backend, x, prompt) {
 }
 
 #' @export
-m_backend_submit.mall_simulate_llm <- function(backend, x, base_prompt) {
+m_backend_submit.mall_simulate_llm <- function(backend, x, prompt) {
   args <- backend
   class(args) <- "list"
   if (args$model == "pipe") {
-    out <- trimws(strsplit(x, "\\|")[[1]][[2]])
-  } else if (args$model == "prompt") {
-    out <- glue("{base_prompt}\n{x}")
+    out <- map_chr(x, \(x) trimws(strsplit(x, "\\|")[[1]][[2]]))
   } else if (args$model == "echo") {
     out <- x
-  } else {
-    out <- list(
-      x = x,
-      base_prompt = base_prompt,
-      backend = args
-    )
   }
   out
 }

R/mall.R

@@ -1,5 +1,5 @@
 #' @importFrom ollamar chat test_connection list_models
-#' @importFrom dplyr mutate tibble bind_cols
+#' @importFrom dplyr mutate tibble bind_cols pull
 #' @importFrom utils menu
 #' @importFrom jsonlite fromJSON
 #' @import rlang

tests/testthat/_snaps/llm-classify.md

@@ -0,0 +1,44 @@
+# Classify on Ollama works
+
+    Code
+      llm_classify(reviews, review, labels = c("appliance", "computer"))
+    Output
+                                                                                    review
+      1                 This has been the best TV I've ever used. Great screen, and sound.
+      2          I regret buying this laptop. It is too slow and the keyboard is too noisy
+      3 Not sure how to feel about my new washing machine. Great color, but hard to figure
+        .classify
+      1 appliance
+      2  computer
+      3 appliance
+
+---
+
+    Code
+      llm_classify(reviews, review, pred_name = "new", labels = c("appliance",
+        "computer"))
+    Output
+                                                                                    review
+      1                 This has been the best TV I've ever used. Great screen, and sound.
+      2          I regret buying this laptop. It is too slow and the keyboard is too noisy
+      3 Not sure how to feel about my new washing machine. Great color, but hard to figure
+              new
+      1 appliance
+      2  computer
+      3 appliance
+
+---
+
+    Code
+      llm_classify(reviews, review, pred_name = "new", labels = c("appliance",
+        "computer"), additional_prompt = "Consider all laptops as appliances.")
+    Output
+                                                                                    review
+      1                 This has been the best TV I've ever used. Great screen, and sound.
+      2          I regret buying this laptop. It is too slow and the keyboard is too noisy
+      3 Not sure how to feel about my new washing machine. Great color, but hard to figure
+              new
+      1 appliance
+      2 appliance
+      3 appliance
+

tests/testthat/_snaps/llm-custom.md

@@ -0,0 +1,14 @@
+# Custom on Ollama works
+
+    Code
+      llm_custom(reviews_table(), review, my_prompt)
+    Output
+                                                                                    review
+      1                 This has been the best TV I've ever used. Great screen, and sound.
+      2          I regret buying this laptop. It is too slow and the keyboard is too noisy
+      3 Not sure how to feel about my new washing machine. Great color, but hard to figure
+        .pred
+      1   Yes
+      2    No
+      3    No
+

tests/testthat/_snaps/llm-extract.md

@@ -0,0 +1,14 @@
+# Extract on Ollama works
+
+    Code
+      llm_extract(reviews_table(), review, "product")
+    Output
+                                                                                    review
+      1                 This has been the best TV I've ever used. Great screen, and sound.
+      2          I regret buying this laptop. It is too slow and the keyboard is too noisy
+      3 Not sure how to feel about my new washing machine. Great color, but hard to figure
+               .extract
+      1              tv
+      2          laptop
+      3 washing machine
+