Don't use assert in non-test code.

freqtrade/freqai/torch/PyTorchModelTrainer.py

@@ -148,7 +148,8 @@ class PyTorchModelTrainer(PyTorchTrainerInterface):
         the motivation here is that `n_steps` is easier to optimize and keep stable,
         across different n_obs - the number of data points.
         """
-        assert isinstance(self.n_steps, int), "Either `n_steps` or `n_epochs` should be set."
+        if not isinstance(self.n_steps, int):
+            raise ValueError("Either `n_steps` or `n_epochs` should be set.")
         n_batches = n_obs // self.batch_size
         n_epochs = max(self.n_steps // n_batches, 1)
         if n_epochs <= 10: