Revamp tensor constructors.

feiwang3311 / Lantern

BSD 3-Clause "New" or "Revised" License

167 stars 15 forks source link

Closed dan-zheng closed 5 years ago

dan-zheng commented 5 years ago

Make constructors that take both "shape" and "data" arguments:

Specify shape as an explicit Seq[Int], not Int*. This clearly distinguishes shape from data.
Specify shape as the first argument. This is consistent with libraries like NumPy (e.g. np.full(shape, fill_value)).

Rename zeros(Tensor) and ones(Tensor) to zeros_like and ones_like.

dan-zheng commented 5 years ago

Multiple parameter lists is a great idea, noted in #17. I'll continue to focus on GPU support.