This project demonstrates AutoML + Hyperparameter Optimization (HPO) using Optuna for different deep learning architectures (CNN & RNN) on the IMDB movie review dataset.

The project is structured as a proper Python package, making it easy to extend, reuse, and integrate into MLOps pipelines.

deeplearn_pipeline_optimizer/

data/
 imdb_loader.py Loads & preprocesses IMDB dataset 
models/
 cnn_text.py CNN architecture for text classification
 rnn_text.py RNN (GRU) architecture for text classification
optimizers/
 optuna_runner.py Optuna search space & optimization loop
train/
 train_model.py Training & evaluation helper
main.py CLI entry point
requirements.txt
README.md

1. Load & preprocess IMDB data (padding, tokenization).
2. Choose architecture (CNN or RNN).
3. Define hyperparameter search space in optuna_runner.py.
4. Run Optuna optimization to find the best parameters.
5. Train & evaluate using the best configuration.

git clone https://github.com/b-tanyileke/deeplearn_pipeline_optimizer.git
cd deep-learning-pipeline-optimizer
pip install -r requirements.txt

Optimize CNN

• python main.py --model cnn --trials 15

Optimize RNN

• python main.py --model rnn --trials 15

Running Optuna optimization for model: cnn Number of trials: 15

Best trial (cnn): Accuracy: 0.8894 embedding_dim: 128 dropout_rate: 0.3 learning_rate: 0.0005 batch_size: 64 num_filters: 128 kernel_size: 5

Architecture-agnostic optimization — easily plug in more models (e.g., Transformer).

Reusable training pipeline.

Search spaces per architecture (CNN vs RNN have different tunable params).

Early stopping to avoid overfitting during trials.

feel free to use and modify.