This advanced training will illustrate how one can prepare and store data in GreyCat while leveraging Python dedicated libraries to train machine learning models.
- Python >= 3.11
- pip
In this training we use the following dataset:
Patrick Fleith. (2023). Controlled Anomalies Time Series (CATS) Dataset (Version 2) [Data set]. Solenix Engineering GmbH. https://doi.org/10.5281/zenodo.8338435
mkdir -p data
curl -L https://huggingface.co/datasets/patrickfleith/controlled-anomalies-time-series-dataset/resolve/main/data.csv > data/dataset.csvpython3 -m pip install -r requirements.txt.
├── configs # History of config files ever run. Datetime in the filename.
├── config.yaml # Parameters of the execution.
├── data
│ └── dataset.csv
├── lib # Python library used by the script.
│ ├── architecture.py # Definition of the model class (in this case the Transformer).
│ ├── engine.py # Functions required to run the training loop.
│ ├── processing.py # Implementation of the data loader for training each optimization step.
│ └── utils.py # Auxiliary functions.
├── LICENSE.txt
├── main.py # Script to be executed in order to run the experiment.
├── project.gcl # GreyCat file containing the functions to get and insert data in the database.
├── README.md
├── requirements.txt
├── server # GreyCat source files implementing importer and data model.
│ ├── edi
│ │ └── importer.gcl # Function to load the data from the CSV file to the database.
│ └── model
│ └── model.gcl # Implementation of the data model of the database, which comes with a normalize method.
└── trained-models # Folder holding all the models generated as a result of training. Datetime in the filename.
config and trained-models folders are not existing by default, but they are created when needed in case they don't exist.
-
Start the GreyCat server:
greycat serve --user=1
The first time you execute this command, the dataset will be imported and preprocessed in GreyCat.
- By default the disk storage is limited to 10240MB. If one wants to import the whole dataset it might be necessary to raise this limit:
It is important to notice that on Windows systems, a space of this size is fully reserved as soon as GreyCat starts.
echo 'GREYCAT_STORE=16384' > .env
- By default the disk storage is limited to 10240MB. If one wants to import the whole dataset it might be necessary to raise this limit:
-
All the parameters of the execution are in the
config.yamlfile; in order to run the script:python3 -m main
dataset_id: (int) # id of the dataset in the Greycat database.
dataloader:
n_features: (int) # number of features to use when training (must be multiple of TransformerAutoencoder.n_heads).
n_rows: (int) # number of rows to load for training.
batch_size: (int) # number of sequences to be forward and back-propagated in parallel in each training iteration.
window_len: (int) # number of time steps of each sequence.
delay: (int) # difference in time steps between the input and target output sequence.
TransformerAutoencoder:
n_heads: (int) # number of heads of the attention layers (must be divisor of dataloader.n_features).
n_encoder_layers: (int) # number of layers (Attention + Feed Forward) of the encoding part.
n_decoder_layers: (int) # number of layers (Attention + Feed Forward) of the decoding part.
model_selection:
model_lib: (string) # name of the python module where the model architecture is defined.
model_name: (string) # name of the model architecture class.
training:
device: (string) # choose between cpu or gpu.
learning_rate: (float) # learning rate of the optimizer.
epochs: (int) # number of times all the dataset will be sweeped.
val_size: (float) # between 0 and 1, the proportion of the dataset reserved for validate the results.
save_model: (bool) # whether you want to save results and trained model, and push the information back to GreyCat.