todo.tasks

TODO:
    ☐ Borrow global state so that you never can forget to set it. @high
    ☐ Make `B.jit` work with Torch and TF generators to ensure uniform patterns. @high

    ☐ Check optimality of `move_to_device`. @low

    ☐ Reuse Plum's error message.

Bugs:
    ☐ Is broadcasting of shapes in `B.randgamma` safe with the JIT? @high

Functions:
    ☐ eigvals
    ☐ norm
    ☐ dot
    ☐ cos_sim

＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿
Archive:
 ✓ Allow to index with `int32` for Torch @high @done (22-04-28 15:50) @project(TODO)
 ✓ Add test like this: @high @done (22-04-28 15:50) @project(TODO)
  import lab as B
  import tensorflow as tf
  import lab.tensorflow
  import jax.numpy as jnp
  import lab.jax
  import torch
  import lab.torch
  for dtype in [np.float32, jnp.float32, tf.float32, torch.float64]
 ✓ Let `cholesky_solve` for PyTorch use `torch.cholesky_solve` once the derivative is implemented. @done (22-03-30 19:33) @project(Future)
 ✓ Jax @done (22-03-30 19:32) @project(TODO / Support)
 ✓ Design with AutoGrad as well? @done (22-03-30 19:32) @project(TODO / Support)
 x Refactor tests to use PyTest: remove raises, fixtures, and parametrisation. @high @cancelled (22-03-30 19:32) @project(Functions)
 x Refactor scan once TF2.0 is integrated. @cancelled (19-07-07 18:53) @project(TODO)
 ✓ Port bvn_cdf. @done (19-05-16 18:06) @project(TODO)
 ✓ Check Python 2 and Python 3 compatibility. @done (19-05-16 18:06) @project(TODO)
 ✓ Documentation. @critical @done (19-05-02 13:29) @project(TODO)
 ✓ Add support for AutoGrad. @done (19-05-01 13:25) @project(TODO)
 x = B.range(dtype, 10) @project(TODO)
  perm = B.randperm(B.dtype_int(dtype), 10)
  B.take(x, B.cast(B.promote_dtypes(B.dtype_int(dtype), int), perm))