Assistant for predmicror — predmicror

Answer questions about predmicror models using a deterministic local registry and, when available, an optional local Ollama model for prose. Model-fitting code is generated from package metadata and statically checked before being returned.

Usage

predmicror_assistant(
  query,
  model = "llama3-groq-tool-use:8b",
  root = NULL,
  data = NULL,
  file = NULL,
  sheet = NULL,
  sep = NULL,
  dec = ".",
  na.strings = c("", "NA"),
  task = NULL,
  time = NULL,
  response = NULL,
  temperature = NULL,
  ph = NULL,
  aw = NULL,
  inhibitor = NULL,
  return_context = FALSE,
  conversation = NULL,
  backend = c("auto", "ollama", "deterministic"),
  prefer_wrappers = TRUE,
  verify_code = TRUE,
  return_trace = FALSE
)

Arguments

query: Character question to ask.
model: Ollama model name used when backend is "auto" or "ollama".
root: Path to the package root for context collection. Defaults to the installed package path when available, otherwise the current working directory.
data: Optional data frame to profile and use for data-aware code generation.
file: Optional path to a .csv, .txt, .tsv, .xls, or .xlsx file to read, profile, and use for data-aware code generation.
sheet: Optional Excel sheet name or index when file is .xls or .xlsx.
sep: Optional field separator for delimited text files. If NULL, a simple separator detector is used.
dec: Decimal mark for delimited text files.
na.strings: Character vector of strings to treat as missing values when reading delimited text files.
task: Optional data task override. Use one of "growth", "inactivation", or "cardinal"; NULL keeps automatic detection.
time, response, temperature, ph, aw, inhibitor: Optional column-name overrides used when data or file is supplied.
return_context: Logical; if TRUE, returns a list with answer and context.
conversation: Optional list or character vector with prior questions and answers to include as conversational context.
backend: Character string. One of "auto", "ollama", or "deterministic". "auto" uses Ollama when the CLI is available and otherwise falls back to deterministic registry-based output.
prefer_wrappers: Logical; if TRUE, generated fitting examples prefer fit_growth(), fit_inactivation(), and fit_cardinal() over direct gslnls::gsl_nls() calls.
verify_code: Logical; if TRUE, statically checks generated R code with parse() and simple model-scale/signature rules.
return_trace: Logical; if TRUE, returns prompt, backend, candidates, generated code, and validation metadata for debugging.

Value

Character response by default; list with answer and context when return_context = TRUE or list with trace when return_trace = TRUE.

Examples

predmicror_assistant("How do I fit a Huang model?", backend = "deterministic")
#> [1] "\nSuggested model: `HuangFM` - full Huang primary growth model.\nExpected response scale: `lnN`; example response column: `lnN`.\nI used `fit_growth()` as the main path; direct `gsl_nls()` calls are better left for advanced use.\nGrowth responses should be on the natural logarithm scale (lnN).\n\n```r\ndata(growthfull)\nfit <- fit_growth(\n  data = growthfull,\n  model = \"HuangFM\",\n  time = \"Time\",\n  response = \"lnN\",\n  start = list(Y0 = 0, Ymax = 22, MUmax = 1.7, lag = 5)\n)\nsummary(fit)\nfit_metrics(fit)\naug <- predmicror_augment(fit)\nnew_times <- data.frame(Time = seq(min(growthfull$Time), max(growthfull$Time), length.out = 200))\npred <- predict(fit, newdata = new_times)\nplot(lnN ~ Time, data = growthfull, xlab = \"Time\", ylab = \"ln N\")\nlines(new_times$Time, pred, col = \"blue\", lwd = 2)\npoints(growthfull$Time, aug$.fitted, pch = 16, col = \"grey40\")\n```\nVerification: the code passed static parse() and simple scale/signature checks."