When to set shape for Input data?

[EmbraceLife]

Data is preprocessed (shape, norm, one-hot) in Data supplier, and then made batch through data provider. Then it seems that data is well prepared for training.

Then why do we ever need to change the shape of Input data inside Model: - input:?

model:
  - input:
      # when and how should I change data shape?
      shape: 100 #[28,28,1] # 100
    name: images

File "/Users/Natsume/Documents/kur/kur/containers/layers/placeholder.py", line 191, in _build self._infer_shape(model) File "/Users/Natsume/Documents/kur/kur/containers/layers/placeholder.py", line 139, in _infer_shape .format(self.name, self._shape, inferred_shape)) ValueError: Placeholder "images" does not have the same dimensionality as the data source it is connected to. Placeholder: (100,), data source: (28, 28, 1).

[EmbraceLife]

My guess:

• most cases, we are certain about the shape of data we want, so all the preprocessing is done inside data supplier
• Then before training, we put data samples into batches through data provider
• I don't see the use of changing data shape in Model: - input: shape