分类问题通用架构

😋😝🥸🫠😯

数据预处理

将原始波形数据转化为语谱图（process_code）

训练

nohup python src/run/run.py > nohups/training.log 2>&1 &

参数说明: args_log

验证集结果

id	loss	_accuracy	precision_weighted	precision_macro	precision_micro	recall_weighted	recall_macro	recall_micro	f1_weighted	f1_macro	f1_micro	_seed	criterion_name	classifier_params_input_dim	classifier_params_dropout_rate	learning_rate_visual	learning_rate_other	weight_decay_visual	weight_decay_other	_batch_size
vgg19	0.066413604	0.972678399	0.971404303	0.958877132	0.970452447	0.970452447	0.954976741	0.970452447	0.970439361	0.956126703	0.970452447	1	ASLSingleLabel	25088	0.1	0.0001	0.007	0.004	0.002	256
shufflenetv2	0.095507267	0.967274396	0.968572354	0.953265232	0.966874513	0.966874513	0.94848694	0.966874513	0.967098648	0.949349401	0.966874513	1	ASLSingleLabel	50176	0.1	0.0003	0.004	0.001	0.004	32
resnet50	0.098049748	0.966242757	0.969544823	0.955430396	0.968546638	0.968546638	0.961476312	0.968546638	0.96858964	0.957352674	0.968546638	1	ASLSingleLabel	2048	0.1	0.0001	0.001	0.01	0.005	256
resnet18	0.114296517	0.966531853	0.968925758	0.959056621	0.968185105	0.968185105	0.948541479	0.968185105	0.968102855	0.951582671	0.968185105	1	ASLSingleLabel	512	0.1	0.0001	0.003	0.004	0.003	256

预测结果

1. 按照格式要求进行预测: prediction_submit

   nohup python src/run/test_submit.py > nohups/predictions.log 2>&1 &

2. 训练集 + 验证集: train_predictions
3. 测试集: test_predictions