在训练中输出的日志可以判断模型的情况吗?比如过拟合
Created by: yeyupiaoling
- 在分类训练中,通常会输出这样的日志:
Pass 0, Batch 0, Cost 28.375774, {'classification_error_evaluator': 0.703125}
......................
Test with Pass 0, {'classification_error_evaluator': 0.6833333373069763}
Pass 1, Batch 0, Cost 1.136431, {'classification_error_evaluator': 0.6796875}
......................
Test with Pass 1, {'classification_error_evaluator': 0.6333333253860474}
Pass 2, Batch 0, Cost 1.088741, {'classification_error_evaluator': 0.59375}
......................
Test with Pass 2, {'classification_error_evaluator': 0.6333333253860474}
Pass 3, Batch 0, Cost 1.109857, {'classification_error_evaluator': 0.6484375}
......................
Test with Pass 3, {'classification_error_evaluator': 0.6666666865348816}
Pass 4, Batch 0, Cost 1.084528, {'classification_error_evaluator': 0.625}
......................
Test with Pass 4, {'classification_error_evaluator': 0.6333333253860474}
Pass 5, Batch 0, Cost 1.084450, {'classification_error_evaluator': 0.59375}
......................
Test with Pass 5, {'classification_error_evaluator': 0.6666666865348816}
Pass 6, Batch 0, Cost 1.103440, {'classification_error_evaluator': 0.6953125}
......................
Test with Pass 6, {'classification_error_evaluator': 0.6333333253860474}
Pass 7, Batch 0, Cost 1.092799, {'classification_error_evaluator': 0.578125}
......................
Test with Pass 7, {'classification_error_evaluator': 0.6000000238418579}
Pass 8, Batch 0, Cost 1.099837, {'classification_error_evaluator': 0.6796875}
......................
Test with Pass 8, {'classification_error_evaluator': 0.5333333611488342}
Pass 9, Batch 0, Cost 1.095620, {'classification_error_evaluator': 0.6796875}
......................
Test with Pass 9, {'classification_error_evaluator': 0.6333333253860474}
- 在这些日志中可以判断model有什么问题吗?