require user to input Deploy( ... metric ...)

[mikkokotila]

Related with #282

[PiaCuk]

Also I would like to report that you have to spell the metric's name out for Deploy() to work, otherwise you get a KeyError. Here is an example:

define model(x, y, x_val, y_val, params):
    [...]
    optimizer = optimizers.adam(lr_normalizer(params['lr'], optimizers.adam))
    model.compile(loss=params['losses'], optimizer=optimizer, metrics=['mae'])
    
    out = model.fit_generator(generator=train_generator, validation_data=val_generator,
                              epochs=params['epochs'], callbacks=[ta.live()])

    return out, model

dummy_x = np.empty((1, _channels))
dummy_y = np.empty((1, _channels))
scan = ta.Scan(x=dummy_x,
               y=dummy_y,
               model=model,
               params=params,
               grid_downsample=0.1,
               dataset_name='example',
               experiment_no='1')

ta.Deploy(scan, 'example', metric='mae', asc=True)

Traceback (most recent call last):
    [...]
KeyError: 'mae'

It works if you write

ta.Deploy(scan, 'example', metric='val_mean_absolute_error', asc=True)

[BraveDistribution]

I can make a PR @mikkokotila for this issue, if needed.

[mikkokotila]

This is now fixed in v.0.6.1 and currently available in pip install git+https://github.com/autonomio/talos@daily-dev.

Closing here.