1. In the case of Sequential,
if you want to specify the size of the batch, you need to
use batch_input_shape in the input shape of the first layer instead of input_shape, because input_shape cannot specify the size of the batch, and the batch can only be None
input_shape and batch_input_shape.
input_shape does not include the batch size,
batch_input_shape is the shape of the whole input, including the batch size.
2. In the case of functional
Input parameter
shape: a size tuple (integer), excluding batch size. A shape tuple (integer), not including the batch size. For example, shape=(32,) indicates that the expected input is a 32-dimensional vector in batches.
batch_shape: A size tuple (integer) containing the batch size. For example, batch_shape=(10, 32) indicates that the expected input is 10 32-dimensional vectors. batch_shape=(None, 32) indicates a 32-dimensional vector of any batch size.
The explanation for the official website is like this
# 作为 Sequential 模型的第一层
model = Sequential()
model.add(Dense(32, input_shape=(16,)))
# 现在模型就会以尺寸为 (*, 16) 的数组作为输入，
# 其输出数组的尺寸为 (*, 32)
# 在第一层之后，你就不再需要指定输入的尺寸了：
model.add(Dense(32))