Python mxnet回归除以零误差

尝试编辑的示例，以开发用于求解二次方程的机器学习解决方案。我现在得到被零除的错误：

INFO:root:Epoch[0] Train-mse=49.961319
INFO:root:Epoch[0] Time cost=0.030
INFO:root:Epoch[0] Validation-mse=58229.367065
INFO:root:Epoch[1] Batch [2]    Speed: 2000.14 samples/sec  mse=361.597036
INFO:root:Epoch[1] Batch [4]    Speed: 2000.14 samples/sec  mse=1903.920013
INFO:root:Epoch[1] Batch [6]    Speed: 2000.14 samples/sec  mse=6117.729675
INFO:root:Epoch[1] Batch [8]    Speed: 1999.67 samples/sec  mse=4203.171875
INFO:root:Epoch[1] Batch [10]   Speed: 2000.14 samples/sec  mse=31765.921204
INFO:root:Epoch[1] Batch [12]   Speed: 2000.14 samples/sec  mse=6946.003112
Traceback (most recent call last):  File "C:\Users\ibraheem\.vscode\extensions\ms-python.python-0.9.1\pythonFiles\PythonTools\visualstudio_py_launcher_nodebug.py", line 74, in run
    _vspu.exec_file(file, globals_obj)
  File "C:\Users\ibraheem\.vscode\extensions\ms-python.python-0.9.1\pythonFiles\PythonTools\visualstudio_py_util.py", line 119, in exec_file

    exec_code(code, file, global_variables)
  File "C:\Users\ibraheem\.vscode\extensions\ms-python.python-0.9.1\pythonFiles\PythonTools\visualstudio_py_util.py", line 95, in exec_code
    exec(code_obj, global_variables)
  File "c:\Users\ibraheem\Desktop\OtherProjects\python_AI_ML\Untitled-1.py", line 37, in <module>
    batch_end_callback = mx.callback.Speedometer(batch_size, 2))
  File "C:\Python27amd64\lib\site-packages\mxnet\module\base_module.py", line 506, in fit
    callback(batch_end_params)
  File "C:\Python27amd64\lib\site-packages\mxnet\callback.py", line 159, in __call__
    speed = self.frequent * self.batch_size / (time.time() - self.tic)
ZeroDivisionError: float division by zero

---

我将列车数据的2因子相乘，以避免方程的非实数解

我找到了问题的原因。将

车速表

参数的值从

2

增加到

50

。速度较慢，但仍能正常工作：

model.fit(train_iter, eval_iter,
            optimizer_params={'learning_rate':0.005, 'momentum': 0.9},
            num_epoch=50,
            eval_metric='mse',
            batch_end_callback = mx.callback.Speedometer(batch_size, 50))

---

如果您发布您的code@AndreaCorbellini谢谢。把代码放清楚，它不是“慢”，而是输出统计数据的频率较低。网络培训的速度也一样快。

model.fit(train_iter, eval_iter,
            optimizer_params={'learning_rate':0.005, 'momentum': 0.9},
            num_epoch=50,
            eval_metric='mse',
            batch_end_callback = mx.callback.Speedometer(batch_size, 50))