Frequently Asked Questions¶
Installation & Setup¶
Q: What Python version do I need?¶
A: Python 3.8 or higher. Check with python --version.
Q: Can I use IOValence on Windows/Mac/Linux?¶
A: Yes! IOValence works on all major operating systems.
Q: Do I need GPU?¶
A: No, GPU is optional. CPU works fine, just slower. Install GPU support with pip install iovalence[gpu].
Getting Started¶
Q: How much training data do I need?¶
A: Minimum 100 examples, but 500-1000 is better. More data = better accuracy.
Q: How long does training take?¶
A: Depends on data size and model complexity. Typical: 5-30 minutes on CPU, <5 minutes on GPU.
Q: Can I train multiple agents?¶
A: Yes! Each agent is independent. Create multiple projects.
Models & Training¶
Q: What's the best learning rate?¶
A: Start with 0.001. Adjust based on results. Lower = safer but slower.
Q: How do I know if my model is overfitting?¶
A: Compare training and validation accuracy: - Training: 95%, Validation: 70% = Overfitting - Solution: More data, add dropout, reduce model size
Q: Should I use all my data for training?¶
A: No! Split it: - 70-80% for training - 10-15% for validation - 10-15% for testing
Q: How many epochs should I use?¶
A: Start with 20-30. Use early stopping to prevent overfitting.
Data¶
Q: What data formats does IOValence support?¶
A: CSV, JSON, Excel, parquet. CSV is easiest.
Q: Can I use images/text/numbers together?¶
A: Yes! IOValence handles mixed data types.
Q: How do I handle missing values?¶
A: Fill them or remove rows:
df.fillna(df.mean(), inplace=True) # Fill with mean
df.dropna(inplace=True) # Remove rows
Q: What if my classes are imbalanced?¶
A: Use resampling:
from imblearn.over_sampling import SMOTE
X, y = SMOTE().fit_resample(X, y)
Prediction & Deployment¶
Q: How do I make predictions?¶
A: Use the trained model:
iovalence predict --model model.pkl --input "text here"
Q: Can I deploy to the cloud?¶
A: Yes! AWS, GCP, Azure all supported.
Q: How do I update my model?¶
A: Retrain with new data, then deploy new version.
Q: Can I use my model in an app?¶
A: Yes! Deploy as REST API and call from your app.
Troubleshooting¶
Q: Training is very slow. How do I speed it up?¶
A: Use GPU: pip install iovalence[gpu]
Or reduce batch size and data size.
Q: Model accuracy is poor. What can I do?¶
A: 1. Check data quality 2. Add more training examples 3. Increase training time (epochs) 4. Adjust learning rate 5. Try larger model
Q: "Out of memory" error¶
A: Reduce batch_size in config:
batch_size: 16 # Instead of 64
Q: Model works in training but not in production¶
A: Ensure same preprocessing: - Same text cleaning - Same data scaling - Same categorical encoding
Features & Capabilities¶
Q: Can I use pre-trained models?¶
A: Yes! IOValence supports transfer learning.
Q: Can I combine multiple models?¶
A: Yes, with ensemble methods: - Voting ensemble - Stacking - Blending
Q: Can I export my model?¶
A: Yes! Export to: - Python pickle - ONNX format - SavedModel - TorchScript
Q: Can I monitor model performance?¶
A: Yes! After deployment:
iovalence monitor --model deployed-model
Support & Help¶
Q: Where can I get help?¶
A: - Documentation: Our guide - Issues: GitHub Issues - Forum: Community Forum
Q: Can I contribute?¶
A: Yes! See Contributing Guide
Q: Is IOValence free?¶
A: Open-source version is free. Enterprise version available for large-scale use.
Advanced Questions¶
Q: Can I use custom loss functions?¶
A: Yes, in advanced configuration.
Q: How do I implement custom metrics?¶
A: See Custom Training Pipeline for advanced customization
Q: Can I use distributed training?¶
A: Yes, for large datasets.
Still have questions? Contact Support → or check Common Issues
Q: Can I use IOValence without a GPU?¶
A: Yes, but GPU support significantly speeds up training. CPU-only mode is available.
Installation Questions¶
Q: How do I install IOValence?¶
A: Use pip install iovalence or see Installation Guide.
Q: What if installation fails?¶
A: Check Troubleshooting or run system verification with verify_environment().
Training Questions¶
Q: How long does training take?¶
A: Depends on model size, dataset size, and hardware. Start with a small subset to estimate.
Q: Can I resume training from a checkpoint?¶
A: Yes, use trainer.load_checkpoint() and set resume=True.
Q: How do I prevent overfitting?¶
A: Use dropout, early stopping, data augmentation, and regularization techniques.
Model Questions¶
Q: What model types are supported?¶
A: Transformer, CNN, RNN, LSTM, GRU, and GNN. See Models for details.
Q: Can I use pre-trained models?¶
A: Yes, use Model.from_pretrained() for transfer learning.
Q: How do I create custom models?¶
A: Extend the Agent class and implement your architecture. See Custom Training Pipeline.
Data Questions¶
Q: What data formats are supported?¶
A: Images (PNG, JPEG), text (TXT, CSV), tabular (CSV, Parquet), and audio (WAV, MP3).
Q: How do I handle imbalanced datasets?¶
A: Use class weights, data augmentation, or specialized sampling strategies.
Q: Can I use custom data loaders?¶
A: Yes, IOValence accepts any PyTorch-compatible DataLoader.
Deployment Questions¶
Q: How do I deploy a trained model?¶
A: Use REST API deployment, Docker containers, or Kubernetes. See Deployment Guide.
Q: Can I serve multiple models?¶
A: Yes, use the ModelServer with multiple agent instances.
Performance Questions¶
Q: Why is my model slow?¶
A: Check data loading bottlenecks, enable mixed precision, use batch normalization.
Q: How can I improve training speed?¶
A: Use distributed training, increase batch size, or optimize data preprocessing.
Q: How do I monitor GPU usage?¶
A: Use nvidia-smi or enable monitoring in the trainer configuration.