Args: code_dir: the code directory use_git: if using git, then upload all files that are not ignored by git.

Duplicate global (and optionally level) features across all layers Args: level_policy: If 'drop', level features are sim

Args: input_dim must either be an int, i.e. the expected 1D input tensor dim, or a dict s.t. input_dim and spatia

Args: exp_type (str): 'pristine' or 'clear-sky' target_type (str): 'longwave" or 'shortwave' target_varia

Args: adj: a (N, N) adjacency matrix, where N is the number of nodes Returns: A (2, E) array, edge_idxs, where E is the numbe

Function to be called to apply a torch function to all tensors of this class, e.g. apply .to(), .cuda(), ..., Just call this function

How to transform dict X = { 'layer': layer_array, # shape (batch-size, #layers, #layer-features)

A couple of optional utilities, controlled by main config file: - disabling warnings - easier access to debug mode - forcing debug friendl

Downstream model forward pass, input X will be the (batched) output from self.input_transform

input: Dict with key-values {GLOBALS: x_glob, LEVELS: x_lev, LAYERS: x_lay}, where x_*** are the corresponding featu

input: Dict with key-values {GLOBALS: x_glob, LEVELS: x_lev, LAYERS: x_lay}, where x_*** are the corresponding featu

input: Dict with key-values {GLOBALS: x_glob, LEVELS: x_lev, LAYERS: x_lay}, where x_*** are the corresponding featu

Adjacency matrix of a line graph with global node and self-loops