Flask REST service to train and inference RNN model.
Python requirements:
pip install -r requirements.txt
For best performance it is highly recommended to install CUDA.
To use pretrained models, download from drive base_rnn.ckpt, data_normalizer.pkl and paste them to weights folder.
python app.py -h
usage: app.py [-h] [--model_path MODEL_PATH]
[--normalizer_path NORMALIZER_PATH] [--debug DEBUG]
[--port PORT] [--use_gpu USE_GPU] [--hparams HPARAMS]
optional arguments:
-h, --help show this help message and exit
--model_path MODEL_PATH
Path RNN model weights (default:
weights\base_rnn.ckpt)
--normalizer_path NORMALIZER_PATH
Path to sklearn Normalizer (default:
weights\data_normalizer.pkl)
--debug DEBUG Use debug mode (default: True)
--port PORT The port of the webserver. Defaults to `5000`
(default: 5000)
--use_gpu USE_GPU Use gpu for training and inference (default: True)
--hparams HPARAMS Path to model hparams (default:
weights\rnn_hparams.json)
To predict: POST json on path /predict.
request json example:
{"x": [
[1,2,...],
[..],
.
]}
response json example:
{"y": [
1,
2,
.
]}
To train: POST json on path /train.
request json example:
{"x": [
[1,2,...],
[..],
.
],
"y":[
1,
2,
.
]}
response json example:
{"loss": 1} # mape%
More examples in Postman collection rest_rnn_service.postman_collection.json