Inputdim

Inputdim is a term commonly used to denote the dimensionality of the input data that a model processes. In statistics and machine learning, it refers to the number of features or variables describing each observation. The concept is central to how models are configured, since it determines the size of the input layer and the shape of the data that flows through the network.

In terms of representation, a dataset with n_samples observations and input_dim features is typically arranged in a data matrix with shape (n_samples, input_dim). In deep learning frameworks, tensors may use shapes such as (batch_size, input_dim) for feedforward networks, or include additional axes for sequences or channels (for example, (batch_size, sequence_length, input_dim) for recurrent models).

Determination of input_dim depends on the data. For tabular data, it is the number of columns. For images, input_dim can be the number of pixels when the image is flattened (for instance, 28x28 = 784) or the per-step feature size in a sequence of images. In time-series or sequential data, input_dim is the number of features per time step.

Implications include ensuring compatibility between data and model architecture. A mismatch between input_dim and the expected input size of the first layer can cause errors. During model construction, input_dim often appears as a parameter such as the in_features of a linear layer, and it must align with the dataset’s feature count.

See also: feature dimension, input size, dimensionality, batch size, data shapes.