multi_modal_network return values not compatible with policy_opt_tf

[fhoekstra]

In policy_opt_tf, around line 70:

tf_map_generator = self._hyperparams['network_model']
tf_map, fc_vars, last_conv_vars = tf_map_generator(dim_input=self._dO, dim_output=self._dU,
    batch_size=self.batch_size, network_config=self._hyperparams['network_params'])

return values are expected in the form as multi_modal_network_fp gives them: (tf_model_example.py, line 268)

return nnet, fc_vars, last_conv_vars

However, the multi_modal_network in that same file uses a different expression: (line 166)
return TfMap.init_from_lists([nn_input, action, precision], [fc_output], [loss])

which makes policy_opt_tf raise the following error:

*** TypeError: iteration over non-sequence

Running Tensorflow 0.8.0 on Ubuntu 16.04

[kapsl]

I think I could get rid of this error by using Tensorflow 0.5 - 0.7