of other integer types) in `tf.constant`.
*   Make the `gain` argument of convolutional orthogonal initializers
    (`convolutional_delta_orthogonal`, `convolutional_orthogonal_1D`,
    `convolutional_orthogonal_2D`, `convolutional_orthogonal_3D`) have
    consistent behavior with the `tf.initializers.orthogonal` initializer, i.e.
    scale the output l2-norm by `gain` and NOT by `sqrt(gain)`. (Note that these

build:windows --copt=/D_USE_MATH_DEFINES
build:windows --host_copt=/D_USE_MATH_DEFINES

# Windows has a relatively short command line limit, which TF has begun to hit.
# See https://docs.bazel.build/versions/main/windows.html
build:windows --features=compiler_param_file
build:windows --features=archive_param_file

# Speed Windows compile times. Available in VS 16.4 (we are on 16.11). See