add base_score only when needed

[StrikerRUS]

It was quite strange to see global bias in generated code for LightGBM. Also, it simplifies hence speeds up runtime.

[coveralls]

Coverage increased (+0.004%) to 95.255% when pulling bc2f75c on base_score into ec02c72 on master.

[izeigerman]

This is a great idea 👍 Do you think the same logic can be applied to bias and coefficients in case of linear models? I can see how L1-regularized linear models can benefit from this greatly (relatively).

[StrikerRUS]

Of course, awesome proposal! But I have plans to introduce DotProduct expression with a fallback (very similar one to the current linear_to_ast function). I think I will address sparsity problems there.

m2cgen/m2cgen/assemblers/linear.py

Lines 225 to 233 in ec02c72

	def _linear_to_ast(coef, intercept):
	feature_weight_mul_ops = [
	utils.mul(ast.FeatureRef(index), ast.NumVal(value))
	for index, value in enumerate(coef)
	]
	return utils.apply_op_to_expressions(
	ast.BinNumOpType.ADD,
	ast.NumVal(intercept),
	*feature_weight_mul_ops)

[izeigerman]

Awesome, thank you 👍