Encoder/Decoder - 2d function regression

[l8git]

The 2d x_value generation works for the generation of train_data, but not for test_data (still 1d).

[l8git]

The issue is coming from the Colab notebook, since the data generation is designed for 1d regression. In particular,
it arises from:

def generate_curves(self):

      num_context = tf.random_uniform(shape=[], minval=3, maxval=self._max_num_context, dtype=tf.int32)

      if self._testing:
        num_target = 400
        num_total_points = num_target
        x_values = tf.tile(
            tf.expand_dims(tf.range(-2., 2., 1. / 100, dtype=tf.float32), axis=0),
            [self._batch_size, 1])
        x_values = tf.expand_dims(x_values, axis=-1)

For 2d regression (MNIST) adaptions are necessary. For the encoder we have to make sure that it's accepting dynamic inputs in terms of dimensionality.