*   Raw TensorFlow functions can now be used in conjunction with the Keras
        Functional API during model creation. This obviates the need for users
        to create Lambda layers in most cases when using the Functional API.
        Like Lambda layers, TensorFlow functions that result in Variable
        creation or assign ops are not supported.
    *   Add support for passing list of lists to the `metrics` argument in Keras

* Make Travis run Ubuntu Xenial (newer version) and Python 3.7 instead of Python 3.7-dev. PR [#92](https://github.com/tiangolo/fastapi/pull/92) by [@blueyed](https://github.com/blueyed).

* Fix duplicated param variable creation. PR [#123](https://github.com/tiangolo/fastapi/pull/123) by [@yihuang](https://github.com/yihuang).

0, Q, ..., 8Q v.points[0].FromP3(q) tmpP3 := Point{} tmpP1xP1 := projP1xP1{} for i := 0; i < 7; i++ { // Compute (i+1)*Q as Q + i*Q and convert to a projCached // This is needlessly complicated because the API has explicit // receivers instead of creating stack objects and relying on RVO v.points[i+1].FromP3(tmpP3.fromP1xP1(tmpP1xP1.Add(q, &v.points[i]))) } } // This is not optimised for speed; fixed-base tables should be precomputed. func (v *affineLookupTable) FromP3(q *Point) { // Goal: v.points[i]...

0, Q, ..., 8Q v.points[0].FromP3(q) tmpP3 := Point{} tmpP1xP1 := projP1xP1{} for i := 0; i < 7; i++ { // Compute (i+1)*Q as Q + i*Q and convert to a projCached // This is needlessly complicated because the API has explicit // receivers instead of creating stack objects and relying on RVO v.points[i+1].FromP3(tmpP3.fromP1xP1(tmpP1xP1.Add(q, &v.points[i]))) } } // This is not optimised for speed; fixed-base tables should be precomputed. func (v *affineLookupTable) FromP3(q *Point) { // Goal: v.points[i]...