## ----setup, include=FALSE-----------------------------------------------------
knitr::opts_chunk$set(echo = TRUE)
library(reticulate)

# Replace the path below with the path of your Python environment
# Then uncomment the command below:
# Tip: BERTOPICR_VENV should be the folder that contains `pyvenv.cfg`.
# Sys.setenv(
#   BERTOPICR_VENV = "C:/path/to/your/venv",
#   NOT_CRAN = "true"
# )

# 1. Define the libraries you need
required_modules <- c("bertopic", "umap", "hdbscan", "sklearn", "numpy", "sentence_transformers", "torch")

# macOS: if reticulate fails to load Python libraries, run once per session.
if (identical(Sys.info()[["sysname"]], "Darwin")) {
  bertopicr::configure_macos_homebrew_zlib()
}

# Optional: point reticulate at a user-specified virtualenv
venv <- Sys.getenv("BERTOPICR_VENV")
if (nzchar(venv)) {
  venv_cfg <- file.path(venv, "pyvenv.cfg")
  if (file.exists(venv_cfg)) {
    reticulate::use_virtualenv(venv, required = TRUE)
  } else {
    message("Warning: BERTOPICR_VENV does not point to a valid virtualenv: ", venv)
  }
}

# Try to find python, but don't crash if it's missing (e.g. on another user's machine)
if (!reticulate::py_available(initialize = TRUE)) {
  try(reticulate::use_python(Sys.which("python"), required = FALSE), silent = TRUE)
}

# 2. Check if they are installed
python_ready <- tryCatch({
  # Attempt to initialize python and check modules
  py_available(initialize = TRUE) &&
    all(vapply(required_modules, py_module_available, logical(1)))
}, error = function(e) FALSE)

# 3. Only evaluate chunks when Python is ready and NOT_CRAN is set
run_chunks <- python_ready && identical(Sys.getenv("NOT_CRAN"), "true")
knitr::opts_chunk$set(eval = run_chunks)

if (!python_ready) {
  message("Warning: Required Python modules are not available. Vignette code will not run.")
} else {
  message("Python environment ready: ", reticulate::py_config()$python)
  if (!identical(Sys.getenv("NOT_CRAN"), "true")) {
    message("Note: Set NOT_CRAN=true to run Python-dependent chunks locally.")
  }
}

## -----------------------------------------------------------------------------
# library(reticulate)
# library(bertopicr)
# library(readr)
# library(dplyr)

## -----------------------------------------------------------------------------
# reticulate::py_run_string(code = "import torch
# print(torch.cuda.is_available())") # if GPU is available then TRUE else FALSE

## -----------------------------------------------------------------------------
# sample_path <- system.file("extdata", "spiegel_sample.rds", package = "bertopicr")
# df <- read_rds(sample_path)
# docs <- df |> pull(text_clean)

## -----------------------------------------------------------------------------
# topic_model <- train_bertopic_model(
#   docs = docs,
#   top_n_words = 50L, # set integer numbger of top words
#   embedding_model = "Qwen/Qwen3-Embedding-0.6B", # choose your (multilingual) model from huggingface.co
#   embedding_show_progress = TRUE,
#   timestamps = df$date, # set this to NULL if not applicable with your data
#   classes = df$genre, # set this to NULL if not applicable with your data
#   representation_model = "keybert" # keyword generation for each topic
# )

## -----------------------------------------------------------------------------
# save_bertopic_model(topic_model, "topic_model")