def __init__()

in scripts/datasets.py [0:0]

• 26 lines of code
• 5 McCabe index (conditional complexity)

    def __init__(self, dim_inv, dim_spu, n_envs):
        self.scramble = torch.eye(dim_inv + dim_spu)
        self.dim_inv = dim_inv
        self.dim_spu = dim_spu
        self.dim = dim_inv + dim_spu

        self.task = "classification"
        self.envs = {}

        if n_envs >= 2:
            self.envs = {
                'E0': {"p": 0.95, "s": 0.3},
                'E1': {"p": 0.97, "s": 0.5}
            }
        if n_envs >= 3:
            self.envs["E2"] = {"p": 0.99, "s": 0.7}
        if n_envs > 3:
            for env in range(3, n_envs):
                self.envs["E" + str(env)] = {
                    "p": torch.zeros(1).uniform_(0.9, 1).item(),
                    "s": torch.zeros(1).uniform_(0.3, 0.7).item()
                }

        # foreground is 100x noisier than background
        self.snr_fg = 1e-2
        self.snr_bg = 1

        # foreground (fg) denotes animal (cow / camel)
        cow = torch.ones(1, self.dim_inv)
        self.avg_fg = torch.cat((cow, cow, -cow, -cow))

        # background (bg) denotes context (grass / sand)
        grass = torch.ones(1, self.dim_spu)
        self.avg_bg = torch.cat((grass, -grass, -grass, grass))