Package {llmimpute}

Type:	Package
Title:	Missing Data Imputation via Language Models and Statistics
Version:	0.1.0
Description:	Provides missing data imputation through two complementary engines: a large language model engine that communicates with the 'Anthropic' 'Claude' application programming interface for context-aware semantic imputation, and a fully self-contained offline engine implementing nineteen statistical and machine learning algorithms entirely in base R with no additional package dependencies. Offline methods include mean, median, mode, last observation carried forward, next observation carried backward, hot-deck, predictive mean matching, k-nearest neighbours, ordinary least-squares regression, Lasso with coordinate descent, Ridge with closed-form solution, Bayesian Ridge regression with evidence approximation following MacKay (1992), support vector regression with a radial basis function kernel, classification and regression trees, random forests, gradient boosting, iterative random forest imputation, principal component analysis imputation via iterative singular value decomposition, and nuclear-norm minimisation via singular value thresholding. When no API key is available the package automatically falls back to the offline engine, ensuring full operation in environments without internet access. Every imputed value is accompanied by a confidence score and a plain-language reasoning string, producing reproducible audit trails. The automatic method selector chooses the best algorithm per column based on data type, skewness, missingness rate, and inter-column correlations.
License:	MIT + file LICENSE
Encoding:	UTF-8
RoxygenNote:	7.3.3
Language:	en-US
Config/Needs/check:	spelling
Depends:	R (≥ 4.1.0)
Imports:	httr2 (≥ 1.0.0), methods, jsonlite (≥ 1.8.0), cli (≥ 3.6.0),
Suggests:	testthat (≥ 3.0.0), knitr (≥ 1.40), rmarkdown (≥ 2.14), withr (≥ 2.5.0)
Config/testthat/edition:	3
VignetteBuilder:	knitr
BugReports:	https://cran.r-project.org/submit.html
NeedsCompilation:	no
Packaged:	2026-06-18 06:25:03 UTC; acer
Maintainer:	Sadikul Islam <sadikul.islamiasri@gmail.com>
Author:	Sadikul Islam [aut, cre], Rajesh Kaushal [aut]
Repository:	CRAN
Date/Publication:	2026-06-23 13:40:02 UTC

llmimpute: Missing Data Imputation via Language Models and Statistics

Description

Provides missing data imputation through two complementary engines: a large language model engine that works with any supported language model provider (cloud or local) for context-aware semantic imputation, and a fully self-contained offline engine implementing nineteen statistical and machine learning algorithms entirely in base R with no additional package dependencies.

Supported cloud providers include Anthropic 'Claude', OpenAI, Google Gemini, Groq, Mistral AI, Cohere, OpenRouter, Together AI, Fireworks AI, DeepSeek, Perplexity, xAI Grok, AI21 Labs, and Cerebras. Supported local servers include Ollama, LM Studio, Jan, llama.cpp, KoboldCpp, and Text Generation WebUI. Any other OpenAI-compatible endpoint can be used via the "custom" provider option.

When no API key is available the package automatically falls back to the offline engine, ensuring full operation in environments without internet access.

Quick start

library(llmimpute)

df <- data.frame(
  age    = c(25L, NA, 35L, 40L),
  income = c(50000, 60000, NA, 80000),
  edu    = c("BSc", NA, "MSc", "BSc"),
  stringsAsFactors = FALSE
)

# Offline - no key needed
result <- lmi_impute(df)

# Free cloud (Groq)
# lmi_set_api_key("gsk_...", provider = "groq")

# Free cloud (Gemini)
# lmi_set_api_key("AIza...", provider = "gemini")

# Free local (Ollama)
# lmi_set_api_key(provider = "ollama")

# Free local (LM Studio)
# lmi_set_api_key(provider = "lmstudio")

# Any OpenAI-compatible server
# lmi_set_api_key("key", provider = "custom",
#   base_url = "http://localhost:8000/v1/chat/completions")

Main functions

lmi_providers

List all supported LLM providers.

lmi_set_api_key

Configure provider, key, and URL.

lmi_set_model

Choose the model for LLM imputation.

lmi_diagnose

Inspect missingness without any API call.

lmi_impute

Unified imputation (LLM or offline fallback).

lmi_impute_offline

Offline-only imputation, 19 algorithms.

lmi_models

List recommended models per provider.

lmi_export

Write imputed data and audit trail to files.

Free options

Free cloud: Groq (https://console.groq.com), Google Gemini (https://aistudio.google.com), OpenRouter (https://openrouter.ai), Cerebras (https://cloud.cerebras.ai). Free local (no key): Ollama (https://ollama.com), LM Studio (https://lmstudio.ai), Jan (https://www.jan.ai/).

Author(s)

Maintainer: Sadikul Islam sadikul.islamiasri@gmail.com (ORCID)

Authors:

• Rajesh Kaushal

See Also

Useful links:

• Report bugs at https://cran.r-project.org/submit.html

Call the configured LLM API for imputation

Description

Internal function. Routes the request to the correct provider endpoint using the active protocol (anthropic, openai-compatible, gemini, cohere, or ollama), builds the prompt, and parses the JSON response.

Usage

.lmi_call_api(chunk, domain, confidence, reasoning, flag_suspicious)

Arguments

Value

Named list with elements imputations and suspicious.

Extract the imputed data frame from an lmi_result

Description

Convenience S3 method allowing as.data.frame(result) to extract the imputed data frame from an lmi_result object directly.

Usage

## S3 method for class 'lmi_result'
as.data.frame(x, ...)

Arguments

Value

The imputed data.frame.

Examples

df <- data.frame(x = c(1, NA, 3))
result <- lmi_impute_offline(df, verbose = FALSE)
clean  <- as.data.frame(result)

Diagnose missing data in a data frame

Description

Analyses a data frame and prints a report on missing values, column types, and the number of unique observed values per column. No API calls are made. Use this function before lmi_impute to preview what will be imputed and to choose an appropriate method.

Usage

lmi_diagnose(data, na_strings = NULL)

Arguments

Value

Invisibly returns a data.frame with one row per column and columns column, type, n_missing, pct_missing, n_unique.

See Also

lmi_impute, lmi_impute_offline

Examples

df <- data.frame(
  age    = c(25L, NA, 35L, 40L),
  income = c(50000, 60000, NA, 80000),
  edu    = c("BSc", NA, "MSc", "BSc"),
  stringsAsFactors = FALSE
)
lmi_diagnose(df)
lmi_diagnose(df, na_strings = "N/A")

Export imputed data and audit trail to files

Description

Writes the imputed data frame and the imputation audit trail to disk. Supports CSV (default) and RDS output formats. When flag_suspicious = TRUE was used in lmi_impute, the suspicious-values table is also written.

Usage

lmi_export(
  result,
  path = tempdir(),
  prefix = "llmimpute",
  format = c("csv", "rds"),
  overwrite = FALSE
)

Arguments

Value

Invisibly returns a named character vector of the file paths written.

See Also

lmi_impute, lmi_impute_offline

Examples

df <- data.frame(
  age    = c(25L, NA, 35L),
  income = c(50000, 60000, NA),
  stringsAsFactors = FALSE
)
result <- lmi_impute_offline(df, verbose = FALSE)
## Not run: 
lmi_export(result, path = tempdir(), prefix = "my_study")

## End(Not run)

Impute missing values using LLM or built-in statistical methods

Description

Primary entry point for the llmimpute package. When an Anthropic API key is configured, missing values are filled using the Claude large language model, which reasons about each missing cell using the semantic meaning of column names, inter-column relationships, and domain knowledge. When no API key is available (or when offline = TRUE), the function transparently delegates to lmi_impute_offline, which runs entirely in base R without internet access.

Usage

lmi_impute(
  data,
  domain = c("general", "healthcare", "financial", "hr", "survey", "scientific"),
  offline = FALSE,
  offline_fallback = TRUE,
  offline_method = c("auto", "mean", "median", "mode", "locf", "nocb", "hotdeck", "pmm",
    "knn", "linear", "lasso", "ridge", "bayesian_ridge", "svr", "decision_tree",
    "random_forest", "gradient_boost", "missforest", "pca_impute", "softimpute"),
  na_strings = NULL,
  cols = NULL,
  confidence = TRUE,
  reasoning = TRUE,
  flag_suspicious = FALSE,
  max_rows = 50L,
  knn_k = 5L,
  seed = 42L,
  verbose = TRUE
)

Arguments

Value

An object of class lmi_result, a named list with:

data

The imputed data.frame.

imputations

data.frame audit trail: one row per imputed cell with columns row, col, original, imputed, confidence, reasoning.

suspicious

data.frame of flagged existing values (LLM mode, flag_suspicious = TRUE only). NULL otherwise.

summary

Named list of imputation statistics.

call

The matched call.

Engine selection

See Also

lmi_impute_offline, lmi_diagnose, lmi_export, print.lmi_result

Examples

# Offline imputation (works with no API key)
df <- data.frame(
  age    = c(25L, NA, 35L, 40L, NA),
  income = c(50000, 60000, NA, 80000, 55000),
  edu    = c("BSc", NA, "MSc", "BSc", "PhD"),
  stringsAsFactors = FALSE
)

result <- lmi_impute(df)
result$data
result$imputations
summary(result)

# Force a specific offline method
result2 <- lmi_impute(df, offline = TRUE, offline_method = "random_forest")

## Not run: 
# LLM mode requires a valid Anthropic API key
lmi_set_api_key()   # reads ANTHROPIC_API_KEY from environment
result3 <- lmi_impute(df, domain = "hr")

## End(Not run)

Impute missing values using built-in statistical and ML methods (no API required)

Description

Fully offline imputation using 20 methods from scratch in base R. No API key, no internet, no third-party packages needed.

Usage

lmi_impute_offline(
  data,
  method = c("auto", "mean", "median", "mode", "locf", "nocb", "hotdeck", "pmm", "knn",
    "linear", "lasso", "ridge", "bayesian_ridge", "svr", "decision_tree",
    "random_forest", "gradient_boost", "missforest", "pca_impute", "softimpute"),
  cols = NULL,
  na_strings = NULL,
  knn_k = 5L,
  n_trees = 50L,
  max_depth = 4L,
  shrinkage = 0.1,
  n_iter = 10L,
  lambda = 1,
  rank = 3L,
  seed = 42L,
  verbose = TRUE
)

Arguments

Value

An object of class lmi_result.

See Also

lmi_impute, lmi_methods

Examples

df <- data.frame(
  age    = c(25, NA, 35, 40, NA),
  income = c(50000, 60000, NA, 80000, 55000),
  edu    = c("BSc", NA, "MSc", "BSc", "PhD"),
  stringsAsFactors = FALSE
)
lmi_impute_offline(df)
lmi_impute_offline(df, method = "random_forest")
lmi_impute_offline(df, method = "softimpute")

List all available offline imputation methods

Description

Prints all 20 methods grouped by category with usage guidance.

Usage

lmi_methods()

Value

Invisibly returns a data.frame of method metadata.

Examples

lmi_methods()

List recommended models for each supported LLM provider

Description

Prints recommended model identifiers for every provider supported by lmi_impute, including free-tier options. Use lmi_set_model to activate a model after choosing a provider with lmi_set_api_key.

Usage

lmi_models()

Value

Invisibly returns a character vector of all model identifiers across every provider.

See Also

lmi_set_model, lmi_providers, lmi_impute

Examples

lmi_models()

List all supported LLM providers

Description

Prints every LLM provider supported by lmi_impute, grouped into cloud APIs, local servers, and custom endpoints, with free-tier indicators and required environment variables.

Usage

lmi_providers()

Value

Invisibly returns a character vector of provider names.

See Also

lmi_set_api_key, lmi_models

Examples

lmi_providers()

Configure the API key and LLM provider for llmimpute

Description

Sets the API key and LLM provider used by LLM-mode imputation. Supports cloud APIs (Anthropic 'Claude', OpenAI, Google Gemini, Groq, Mistral, Cohere, OpenRouter, Together AI, Fireworks AI, DeepSeek, Perplexity, xAI Grok, AI21 Labs, Cerebras) and local servers (Ollama, LM Studio, Jan, llama.cpp, KoboldCpp, Text Generation WebUI). Any other OpenAI-compatible endpoint can be used via provider = "custom" with a base_url.

If no API key is configured, lmi_impute automatically falls back to the offline statistical engine. The package is fully functional without any API key.

Usage

lmi_set_api_key(
  api_key = NULL,
  provider = "anthropic",
  base_url = NULL,
  .session = TRUE
)

Arguments

Value

Invisibly returns the API key string (or "" for keyless local providers).

See Also

lmi_providers, lmi_set_model, lmi_impute

Examples

## Not run: 
## Cloud APIs
lmi_set_api_key("sk-ant-...",  provider = "anthropic")
lmi_set_api_key("sk-...",      provider = "openai")
lmi_set_api_key("AIza...",     provider = "gemini")   # free tier
lmi_set_api_key("gsk_...",     provider = "groq")     # free tier
lmi_set_api_key("sk-or-...",   provider = "openrouter") # free models
lmi_set_api_key("...",         provider = "deepseek")
lmi_set_api_key("...",         provider = "cerebras") # free tier

## Local servers (no key needed)
lmi_set_api_key(provider = "ollama")
lmi_set_api_key(provider = "lmstudio")
lmi_set_api_key(provider = "jan")
lmi_set_api_key(provider = "llamacpp")
lmi_set_api_key(provider = "koboldcpp")
lmi_set_api_key(provider = "textgenwebui")

## Any OpenAI-compatible endpoint
lmi_set_api_key("mykey", provider = "custom",
                base_url = "http://my-server:8000/v1/chat/completions")

## Override local port
lmi_set_api_key(provider = "ollama",
                base_url = "http://localhost:11435/api/chat")

## End(Not run)

Set the LLM model used for imputation

Description

Sets or retrieves the model identifier used by lmi_impute in LLM mode. The default model is determined by the active provider. Use lmi_models to see recommended models per provider.

Usage

lmi_set_model(model = NULL)

lmi_get_model()

Arguments

Value

Invisibly returns the active model string.

See Also

lmi_set_api_key, lmi_models

Examples

lmi_get_model()

## Not run: 
lmi_set_model("gpt-4o-mini")
lmi_set_model("gemini-1.5-flash")
lmi_set_model("llama-3.3-70b-versatile")
lmi_set_model("deepseek-chat")
lmi_set_model("llama3.2")        # Ollama local

## End(Not run)

Print an lmi_result object

Description

Displays a formatted summary of an imputation result in the console, including overall statistics, per-column imputation counts, and the first n imputed values with their confidence scores and reasoning.

Usage

## S3 method for class 'lmi_result'
print(x, n = 10L, ...)

Arguments

Value

Invisibly returns x.

Examples

df <- data.frame(
  age    = c(25L, NA, 35L),
  income = c(50000, 60000, NA),
  stringsAsFactors = FALSE
)
result <- lmi_impute_offline(df, verbose = FALSE)
print(result)

Summarise an lmi_result object

Description

Returns a data.frame summarising imputation counts and confidence statistics per column, suitable for further analysis or reporting.

Usage

## S3 method for class 'lmi_result'
summary(object, ...)

Arguments

Value

A data.frame with columns column, n_imputed, mean_confidence, min_confidence, max_confidence. Returns NULL invisibly when no imputations were performed.

Examples

df <- data.frame(
  age    = c(25L, NA, 35L, 40L),
  income = c(50000, 60000, NA, 80000),
  stringsAsFactors = FALSE
)
result <- lmi_impute_offline(df, verbose = FALSE)
summary(result)