A real use case Deep Learning regression model
The app configuration. Describes the neural network hyperparameters, training columns and the target columns to predict. Also defines the normalization and encoding information for some columns.
The main app component. It will create the Keras neural network and initiate the learning process using the sample training dataset in ./dataset/train_dataset.csv.
It will also create a directory ./generated and will add the execution logs along with the model configuration.
Teansorboard will be also available at the end of the execution by typing the command tensorboard --logdir=./generated/tensorboard/<logs-file>.
The command to run it is shown in your terminal or can be found in ./generated/learner/<latest log>
Load the model configuration in ./generated/config/ and makes the predictions using the specified data source.
It will then append the prediction in ./generated/output_data.csv.
The default target column to predict is "age_parent"
Functions librairy used by the learner & predictor and provides data preparation/transformation routines.
Before package installation make sure you are running on Python 3.5+ 64-bit.
Scripts dependencies & installation :
pip install tensorflow
pip install keras
pip install scikit-learn
pip install pandas
pip install numpy
pip install h5py
pip install matplotlib
pip install loguru
To use Tensorflow with GPU support (Recommended) :
pip install tensorflow-gpu
and follow this Guide
python .\app\learner.py .\dataset\train_dataset.csv
data_file Path to the training data file.
--benchmark Compare the mode performance to linear regression.
--feature-importance Evaluate data feature importance.
--kfold-validation Evaluate model performance with kfold validation.
--grid-search Use grid search to find which hyperparameters are the best for the model.
python .\app\learner.py .\dataset\train_dataset.csv --benchmark --feature-importance