TextClassifier example - AttributeError: 'tuple' object has no attribute 'shape'

Hi,

when trying the IMDB Movie Reviews text classification example I receive the following error message when executing

clf.fit(x_train, y_train)

AttributeError: in user code: C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\keras\engine\training.py:803 train_function *\n return step_function(self, iterator)\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\keras\engine\training.py:793 step_function **\n outputs = model.distribute_strategy.run(run_step, args=(data,))\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\distribute\distribute_lib.py:964 run\n return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs)\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\distribute\distribute_lib.py:2315 call_for_each_replica\n return self._call_for_each_replica(fn, args, kwargs)\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\distribute\distribute_lib.py:2674 _call_for_each_replica\n return fn(*args, **kwargs)\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\keras\engine\training.py:786 run_step **\n outputs = model.train_step(data)\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\keras\engine\training.py:756 train_step\n self.compiled_metrics.update_state(y, y_pred, sample_weight)\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\keras\engine\compile_utils.py:389 update_state\n self._build(y_pred, y_true)\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\keras\engine\compile_utils.py:320 _build\n self._metrics, y_true, y_pred)\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\util\nest.py:1131 map_structure_up_to\n **kwargs)\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\util\nest.py:1227 map_structure_with_tuple_paths_up_to\n *flat_value_lists)]\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\util\nest.py:1226 <listcomp>\n results = [func(*args, **kwargs) for args in zip(flat_path_list,\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\util\nest.py:1129 <lambda>\n lambda _, *values: func(*values), # Discards the path arg.\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\keras\engine\compile_utils.py:419 _get_metric_objects\n return [self._get_metric_object(m, y_t, y_p) for m in metrics]\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\keras\engine\compile_utils.py:419 <listcomp>\n return [self._get_metric_object(m, y_t, y_p) for m in metrics]\n C:\Users\Sascha\.conda\envs\keras4\lib\site-packages\tensorflow\python\keras\engine\compile_utils.py:440 _get_metric_object\n y_t_rank = len(y_t.shape.as_list()) AttributeError: ‘tuple’ object has no attribute ‘shape’\n"

I’m using the current tf-nightly 2.3.0.dev20200519 and the current autokeras 1.0.3 on Github. I receive the same error message with tf 2.2 and autokeras 1.0.2, and tf-nightly and autokeras 1.0.2.

I also tried tf 2.1.0 with autokeras 1.0.2, the fit works but then I cannot export/print the best model:

Error in py_call_impl(callable, dots$args, dots$keywords) : AttributeError: ‘TrackableWeightHandler’ object has no attribute ‘shape’

When using tf 2.1.0 with autokeras 1.0.3 I get:

ImportError: cannot import name ‘index_lookup’

upon import of autokeras.

I’m running Python 3.6 on Windows in an isolated conda environment.

Other packages: numpy 1.18.4 pandas 1.0.3 keras-tuner 1.0.1 scikit-learn 0.23.1

Appreciate suggestions on how to get this to run.

[monellz]

I get the same question when I run the example of MNIST!

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.