Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/diazdavilajesus16-stack/IA-LUMINA/llms.txt

Use this file to discover all available pages before exploring further.

Retraining rebuilds the neural network from scratch using the intents defined in respuestas.json. You need to retrain whenever you add new intents, edit existing patterns, or remove outdated entries — any change to the intent data requires a new model before Lumina AI can recognize it.

​
Prerequisites

Make sure the following Python packages are installed before running the training script:
  • tensorflow and keras — neural network architecture and training
  • nltk — tokenization and lemmatization
  • numpy — numerical array operations
Install them with: 
pip install tensorflow nltk numpy

​
Steps

1

Edit respuestas.json

Add or update intent objects in respuestas.json. Each intent needs a tag, a list of patterns, and a list of responses. See the intents configuration guide for the full schema and examples.
2

Run the training script

From the project root directory, execute:
python training_chatbot.py
3

Watch the output

The script prints progress at each stage. You will see the vocabulary count, intent count, training sample count, and final accuracy percentage. If any stage fails, the error message will appear here.
4

Restart the app

After training completes, restart the Streamlit application to load the new model:
streamlit run Main.py

​
What the script produces

Running training_chatbot.py writes three files to the project root:
FileDescription
chatbot_model.h5The trained Keras neural network weights and architecture
words.pklSerialized vocabulary list used for bag-of-words vectorization
classes.pklSerialized list of all intent tag names
These files are loaded at startup by chatbot.py. All three must be present and consistent with each other — they are always regenerated together.

​
Training output example

A successful training run prints the following to the console: 
📖 Procesando intents...
✅ Vocabulario: N palabras | Intents: N
🔧 Generando datos de entrenamiento...
✅ Muestras de entrenamiento: N
🧠 Construyendo modelo...
🚀 Entrenando (esto puede tardar 1-3 minutos)...
✅ ¡Modelo guardado como chatbot_model.h5!
   Accuracy final: XX.XX%
🎉 Listo. Ahora puedes ejecutar: streamlit run main.py
The vocabulary count reflects the unique lemmatized tokens across all patterns. The sample count equals the total number of pattern strings across all intents after shuffling.

​
Training hyperparameters

The model is trained with the following fixed hyperparameters defined in training_chatbot.py:
ParameterValueDescription
epochs300Number of full passes through the training data
batch_size8Samples processed per gradient update
learning_rate0.005SGD optimizer step size (with Nesterov momentum 0.9)
Running training_chatbot.py overwrites the existing chatbot_model.h5, words.pkl, and classes.pkl files without creating a backup. If you need to preserve the current model, copy those three files to a safe location before retraining.

Build docs developers (and LLMs) love

Get started for freeTalk to us