Causal Inference via Sparse Additive Models with Application to Online Advertising