— 1. SETUP: LOAD LIBRARIES —

——————————————————————-

library(doParallel)

## Warning: package 'doParallel' was built under R version 4.4.3

## Loading required package: foreach

## Loading required package: iterators

## Loading required package: parallel

# CRAN limits the number of cores used during package checks
cores <- min(2, parallel::detectCores())
registerDoParallel(cores = cores)

— 2. LOAD AND PREPARE TRAINING DATA —

# Look for the file in the installed package first
csv_path <- system.file("extdata", "tweets.csv", package = "quickSentiment")

# Fallback for when you are building the package locally
if (csv_path == "") {
  csv_path <- "../inst/extdata/tweets.csv"
}
tweets <- read.csv(csv_path)
set.seed(123)

— 3. PREPROCESS THE TEXT —

——————————————————————-

Use the pre_process() function from our package to clean the raw text.

This step is done externally to the main pipeline, allowing you to reuse

the same cleaned text for multiple different models or analyses in the future.

tweets$cleaned_text <- pre_process(tweets$Tweet)
tweets$sentiment = ifelse(tweets$Avg>0,'P','N')

— 4. RUN THE MAIN TRAINING PIPELINE —

——————————————————————-

This is the core of the package. We call the main pipeline() function

to handle the train/test split, vectorization, model training, and evaluation.

result <- pipeline(
  # --- Define the vectorization method ---
  # Options: "bow" (raw counts), "tf" (term frequency), "tfidf"
  vect_method = "tf",
  
  # --- Define the model to train ---
  # Options: "logit", "rf", "xgb"
  model_name = "rf",
  
  # --- Specify the data and column names ---
  df = tweets,
  text_column_name = "cleaned_text",      # The column with our preprocessed text
  sentiment_column_name = "sentiment",    # The column with the target variable
  
  # --- Set vectorization options ---
  # Use n_gram = 2 for unigrams + bigrams, or 1 for just unigrams
  n_gram = 1
)

## --- Running Pipeline: TF + RF ---

## Data split: 946 training rows, 235 test rows.

## Vectorizing with TF (ngram=1)...

##   - Fitting BoW model (tf) on training data...

##   - Applying BoW transformation (tf) to new data...

## 
## --- Training Random Forest Model (with ranger) ---

## Ranger training complete.

===================================================================

— 5. PREDICTION ON NEW, UNSEEN DATA —

===================================================================

The training is complete. The ‘result’ object now contains our trained

model and all the necessary “artifacts” for prediction.

tweets$sentimentPredict <- prediction(
  pipeline_object = result,
  df = tweets,
  text_column = "cleaned_text"
)

## --- Preparing new data for prediction ---

##   - Applying BoW transformation (tf) to new data...

## --- Making Predictions ---

##   - Detected a ranger model. Using predict.train().

## Step 4: Correcting column names for ranger model...

## --- Predictions Complete ---