Transformed target regressor support?

When I build a pipeline where I use a scikeras KerasRegressor, but put it inside of a scikitlearn transformedtargetregressor, I can no longer access the training history.

I can still access the KerasRegressor object, and when I do I get the following:

In [38]: pipeline['NN_model'].regressor
Out[38]: 
KerasRegressor(
	model=<function build_regressor_model at 0x000001DE15A4D5E0>
	build_fn=None
	warm_start=False
	random_state=None
	optimizer=<tensorflow.python.keras.optimizer_v2.adam.Adam object at 0x000001DE760DAF40>
	loss=mse
	metrics=None
	batch_size=None
	validation_batch_size=None
	verbose=1
	callbacks=None
	validation_split=0.2
	shuffle=True
	run_eagerly=False
	epochs=100
	parameters={'num_nodes': (250, 150, 100), 'num_dense_layers': 3, 'dropout': 0.01, 'dropout_type': 'none', 'activation': 'elu', 'output_activation': 'linear', 'learning_rate': 0.0001, 'input_dim': 258, 'output_dim': 140, 'loss': 'mse', 'epochs': 100}
)

but when I try to get the history object, I get the following

In [39]: pipeline['NN_model'].regressor.history
Traceback (most recent call last):

  File "<ipython-input-40-eb7647671dbe>", line 1, in <module>
    pipeline['NN_model'].regressor.history

AttributeError: 'KerasRegressor' object has no attribute 'history'

Am I doing something wrong, or is some of the functionality of scikeras not supported when using a Transformed target regressor?

[stsievert]

Try history_, not history.

https://github.com/adriangb/scikeras/blob/7c072fbed1985e9ad58e2291b3fee2324e4c39ea/scikeras/wrappers.py#L94-L95

[adriangb]

Glad we could help!