Handle all models with error

[pplonski]

The AutoML crashes if all models have error. It should be handled more gently.

The example of crash:

AutoML directory: AutoML_88
The task is multiclass_classification with evaluation metric logloss
AutoML will use algorithms: ['MLP']
AutoML steps: ['simple_algorithms', 'default_algorithms', 'not_so_random', 'hill_climbing_1', 'hill_climbing_2']
Skip simple_algorithms because no parameters were generated.
* Step default_algorithms will try to check up to 1 model
The least populated class in y has only 2 members, which is less than n_splits=5.
There was an error during 1_Default_MLP training.
Please check AutoML_88/errors.md for details.
* Step not_so_random will try to check up to 4 models
There was an error during 1_MLP training.
Please check AutoML_88/errors.md for details.
There was an error during 2_MLP training.
Please check AutoML_88/errors.md for details.
There was an error during 3_MLP training.
Please check AutoML_88/errors.md for details.
There was an error during 4_MLP training.
Please check AutoML_88/errors.md for details.
Traceback (most recent call last):
  File "examples/scripts/nn_benchmark.py", line 53, in <module>
    mlp.fit(train_X, train_y)
  File "/home/piotr/sandbox/mljar-supervised/supervised/automl.py", line 276, in fit
    return self._fit(X, y)
  File "/home/piotr/sandbox/mljar-supervised/supervised/base_automl.py", line 723, in _fit
    raise e
  File "/home/piotr/sandbox/mljar-supervised/supervised/base_automl.py", line 672, in _fit
    step, self._models, self._results_path, self._stacked_models
  File "/home/piotr/sandbox/mljar-supervised/supervised/tuner/mljar_tuner.py", line 105, in generate_params
    return self.get_hill_climbing_params(models)
  File "/home/piotr/sandbox/mljar-supervised/supervised/tuner/mljar_tuner.py", line 335, in get_hill_climbing_params
    unique_model_types = np.unique(df_models.model_type)
  File "/home/piotr/sandbox/mljar-supervised/venv_mljs/lib/python3.6/site-packages/pandas/core/generic.py", line 5136, in __getattr__
    return object.__getattribute__(self, name)
AttributeError: 'DataFrame' object has no attribute 'model_type'

• the MLP algorithm in the example is experimental. It should be Neural Network used instead.