Usage of DeepRenewal

deeprenewal usage

usage: deeprenewal [-h] [--use-cuda USE_CUDA]
                   [--datasource {retail_dataset}]
                   [--regenerate-datasource REGENERATE_DATASOURCE]
                   [--model-save-dir MODEL_SAVE_DIR]
                   [--point-forecast {median,mean}]
                   [--calculate-spec CALCULATE_SPEC]
                   [--batch_size BATCH_SIZE]
                   [--learning-rate LEARNING_RATE]
                   [--max-epochs MAX_EPOCHS]
                   [--number-of-batches-per-epoch NUMBER_OF_BATCHES_PER_EPOCH]
                   [--clip-gradient CLIP_GRADIENT]
                   [--weight-decay WEIGHT_DECAY]
                   [--context-length-multiplier CONTEXT_LENGTH_MULTIPLIER]
                   [--num-layers NUM_LAYERS]
                   [--num-cells NUM_CELLS]
                   [--cell-type CELL_TYPE]
                   [--dropout-rate DROPOUT_RATE]
                   [--use-feat-dynamic-real USE_FEAT_DYNAMIC_REAL]
                   [--use-feat-static-cat USE_FEAT_STATIC_CAT]
                   [--use-feat-static-real USE_FEAT_STATIC_REAL]
                   [--scaling SCALING]
                   [--num-parallel-samples NUM_PARALLEL_SAMPLES]
                   [--num-lags NUM_LAGS]
                   [--forecast-type FORECAST_TYPE]

GluonTS implementation of paper 'Intermittent Demand Forecasting with Deep
Renewal Processes'

optional arguments:
  -h, --help            show this help message and exit
  --use-cuda USE_CUDA
  --datasource {retail_dataset}
  --regenerate-datasource REGENERATE_DATASOURCE
                        Whether to discard locally saved dataset and
                        regenerate from source
  --model-save-dir MODEL_SAVE_DIR
                        Folder to save models
  --point-forecast {median,mean}
                        How to estimate point forecast? Mean or Median
  --calculate-spec CALCULATE_SPEC
                        Whether to calculate SPEC. It is computationally
                        expensive and therefore False by default
  --batch_size BATCH_SIZE
  --learning-rate LEARNING_RATE
  --max-epochs MAX_EPOCHS
  --number-of-batches-per-epoch NUMBER_OF_BATCHES_PER_EPOCH
  --clip-gradient CLIP_GRADIENT
  --weight-decay WEIGHT_DECAY
  --context-length-multiplier CONTEXT_LENGTH_MULTIPLIER
                        If context multipler is 2, context available to hte
                        RNN is 2*prediction length
  --num-layers NUM_LAYERS
  --num-cells NUM_CELLS
  --cell-type CELL_TYPE
  --dropout-rate DROPOUT_RATE
  --use-feat-dynamic-real USE_FEAT_DYNAMIC_REAL
  --use-feat-static-cat USE_FEAT_STATIC_CAT
  --use-feat-static-real USE_FEAT_STATIC_REAL
  --scaling SCALING     Whether to scale targets or not
  --num-parallel-samples NUM_PARALLEL_SAMPLES
  --num-lags NUM_LAGS   Number of lags to be included as feature
  --forecast-type FORECAST_TYPE
                        Defines how the forecast is decoded. For details look
                        at the documentation
An example of training process is as follows:

python3 deeprenewal --datasource retail_dataset --lr 0.001 --epochs 50