A roadmap for regressions problems including:
- feature selection: xgboost importance and correlation of features
- preprocesing: conection with S3 (AWS) to get and preprocess data
- synthetic data generation: Synthetic data generation with VAE's and synthetic minority oversampling techniche (smothe)
- regressions: linear regressions, decission trees, support vector regressor, XGBoost, NN (with and without PCA or Kernel PCA before the training, customs loss functions and activations), CNN architectures, syntethic
- finnetuning: tunner search to find the best architecture on hyper models
Pasos a seguir para comenzar a etiquetar
$ git clone https://github.com/matheus695p/regression-problems-roadmap.git
$ cd regression-problems-roadmap
$ echo instalar los requirements
$ pip install -r requirements.txt
│ .gitignore
│ README.md
│
├───fine_tuning
│ Best_Architectures_resultados.txt
│ best_model.png
│
├───images
│ *.png
├───src
│ ├───cnn_architectures
│ │ cnn_architecture.py
│ │
│ ├───feature_selection
│ │ feature_selection.py
│ │
│ ├───fine_tunning_models
│ │ fine_tuning_models.py
│ │ fine_tuning_module.py
│ │
│ ├───nn_architectures
│ │ main_retrain.py
│ │ module_main.py
│ │ training_module.py
│ │
│ ├───preprocessing
│ │ get_snaphots.py
│ │ get_historical_ads.py
│ │ labeller.py
│ │ labeller_module.py
│ │ preprocessing_historical.py
│ │ preprocessing_snapshots.py
│ │
│ └───synthetic_data
│ autoencoder_synthetic_generation.py
│ generate_synthetic_data.py
├───synthetic_data
│ *.png
└───training_results
Arquitectura 1_ CNN_models.png
Arquitectura 1_ NN_models.png
Arquitectura 1_ PCA_NN_models.png
Arquitectura 2_ CNN_models.png
Arquitectura 2_ NN_models.png
Arquitectura 2_ PCA_NN_models.png
Arquitectura 3_ CNN_models.png
Arquitectura 3_ NN_models.png
Arquitectura 3_ PCA_NN_models.png
Arquitectura 4_ CNN_models.png
Arquitectura 4_ NN_models.png
Arquitectura 4_ PCA_NN_models.png
Arquitectura 5_ CNN_models.png
Arquitectura 5_ NN_models.png
Arquitectura 5_ PCA_NN_models.png
Arquitectura custom loss function_ Final_model.png
CNN_ CNN_final_model.png