scripts

SNNCutoff Scripts

The scripts of training.py and evaluation.py accept various parameters to control different aspects of the training and evaluation.

Content

• Conversion: examples based on conversion algorithm, e.g., QCFS
• DirectTraining: examples based on direction training, e.g., TET and TEBN

Bash Script Example

#!/bin/bash

python ./scripts/training.py \
        base.batch_size=128 \
        base.epochs=300 \
        base.gpu_id='1' \
        base.seed=1200 \
        base.port='11152' \
        base.data='cifar10' \
        base.model='sew_resnet18' \
        base.dataset_path='datasets' \
        \
        snn-train.method='snn' \
        snn-train.arch_conversion=False \
        snn-train.ann_constrs='baseconstrs' \
        snn-train.snn_layers='simplebaselayer' \
        snn-train.regularizer='rcs' \
        snn-train.TET=True \
        snn-train.multistep=True \
        snn-train.add_time_dim=True \
        snn-train.T=6 \
        snn-train.alpha=0.00

Base Parameters

• base.batch_size: Specifies the batch size for training. Default is 128.
• base.epochs: Defines the number of epochs for training. Default is 300.
• base.gpu_id: Indicates the GPU ID to be used for training. Default is '1'.
• base.seed: Sets the random seed for reproducibility. Default is 1200.
• base.port: Port number for distributed training. Default is '11152'.
• base.data: Dataset to be used. For example, cifar10.
• base.model: Model architecture to be used. For example, sew_resnet18, vgg
• base.dataset_path: Path to the dataset directory. Default is 'datasets'.

SNN Training Parameters

• snn-train.method: Training method. For example, 'snn'.
• snn-train.arch_conversion: Boolean to enable or disable architecture conversion. Default is False.
• snn-train.ann_constrs: ANN constraints. For example, 'baseconstrs'.
• snn-train.snn_layers: Specifies the SNN layers. For example, 'simplebaselayer'.
• snn-train.regularizer: Regularizer to be used. For example, 'rcs'.
• snn-train.TET: Boolean to enable or disable TET. Default is True.
• snn-train.multistep: Boolean to enable or disable multi-step training. Default is True.
• snn-train.add_time_dim: Boolean to add a time dimension. Default is True.
• snn-train.T: Time steps for SNN training. Default is 6.
• snn-train.alpha: Alpha parameter for the SNN training. Default is 0.00.

Training and Evaluation

sh scripts/DirectTraining/tet/training.sh

or

sh scripts/DirectTraining/tet/evaluation.sh