Dual Instrumental Variable Regression
Authors: Krikamol Muandet, Arash Mehrjou, Si Kai Lee, Anant Raj
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results show that we are competitive to existing, more complicated algorithms for non-linear instrumental variable regression. Table 1 reports the results of different methods evaluated on the test data. First, we observe that 2SLS achieves the largest MSE in both regimes as expected because the linearity assumption is violated here. |
| Researcher Affiliation | Academia | Krikamol Muandet Max Planck Institute for Intelligent Systems T ubingen, Germany krikamol@tuebingen.mpg.de Arash Mehrjou Max Planck Institute for Intelligent Systems ETH Z urich, Z urich, Switzerland arash.mehrjou@inf.ethz.ch Si Kai Lee Booth School of Business University of Chicago, USA sikai.lee@chicagobooth.edu Anant Raj Max Planck Institute for Intelligent Systems T ubingen, Germany anant.raj@tuebingen.mpg.de |
| Pseudocode | Yes | Algorithm 1 Kernelized Dual IV |
| Open Source Code | Yes | Our implementation is available at https://github.com/krikamol/Dual IV-Neur IPS2020. |
| Open Datasets | Yes | Demand design. We consider the same simulation as in [7, 8]: Y = f(X) + " where Y is outcome, X = (P, T, S) are inputs, and Z = (C, T, S) are instruments. Specifically, Y is sales, P is price, which is endogeneous, C is a supply cost shifter (instrument), and (T, S) are time of year and customer sentiment acting as exogeneous variables. The aim is to estimate the demand function f(p, t, s) = 100+(10+p)s (t) 2p where (t) = 2[(t 5)4/600+exp( 4(t 5)2)+t/10 2]. Training data is sampled according to S Unif{1, . . . , 7}, T Unif[0, 10], (C, V ) N(0, I2), " N( V, 1 2), and P = 25 + (C + 3) (T) + V . |
| Dataset Splits | Yes | Given a dataset (xi, yi, zi)2n i=1 of size 2n, we provide a simple heuristic to determine the values of (λ1, λ2)... To this end, we first estimate ˆf via Proposition 6 and ˆuλ := ( b CW +λI) 1( b CWX ˆf ˆb) on the first half of the data (xi, yi, zi)n i=1. Next, the out-of-sample loss of ˆf is evaluated on the second half (xi, yi, zi)2n i=n+1 using (11) where (yi, zi) are treated as test points. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, memory) are mentioned for running the experiments. |
| Software Dependencies | No | No specific software dependencies with version numbers are provided in the paper. |
| Experiment Setup | Yes | We choose (λ1, λ2) from {10 10, 10 9, . . . , 10 1} via cross-validation. The values of all bandwidth parameters are determined via the median heuristic. |