Conditional Common Entropy for Instrumental Variable Testing and Partial Identification

Authors: Ziwei Jiang, Murat Kocaoglu

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the utility of the proposed method with simulated and real-world datasets. In this section, we first demonstrate the proposed method with simulated data and then provide some case studies with real-world data with instrumental variables.
Researcher Affiliation Academia Ziwei Jiang 1 Murat Kocaoglu 1 1Elmore Family School of Electrical and Computer Engineering, Purdue University. Correspondence to: Ziwei Jiang <jiang622@purdue.edu>.
Pseudocode Yes Algorithm 1 IV Latent Search Input: Joint distribution P(X, Y, Z); Number of iterations N; initialization q(W|X, Y, Z); β0, β1 ≥ 0. for i = 1 to N do Form the joint: qi(X, Y, Z, W) = qi(W|X, Y, Z)P(X, Y, Z). Get posteriors: qi(W) = Σx,y,z qi(X, Y, Z, W) qi(W|X, Y ) = Pz qi(X,Y,Z,W ) / Pz,w qi(X,Y,Z,W ) qi(W|X, Z) = Py qi(X,Y,Z,W ) / Py,w qi(X,Y,Z,W ) Update: qi+1(X, Y, Z, W) = qi(W |X,Z)qi(W |X,Y )qi(U)β0+β1 / f(X,Y,Z)qi(W |X)q(W |Z)β1 where f(X, Y, Z) = Σu qi(W |X,Z)qi(W |X,Y )qi(U)β0+β1 / qi(W |X)q(W |Z)β1 end for Return: q N(W|X, Y, Z)P(X, Y, Z)
Open Source Code Yes Our code is available at https://github.com/ ziwei-jiang/Conditional-Common-Entropy
Open Datasets Yes We first demonstrate the proposed method with simulated data and then provide some case studies with real-world data with instrumental variables. In this section, we demonstrate our result in a more realistic setting with a synthetic dataset introduced by Lauritzen & Spiegelhalter (1988). We provide another example with the Pima Indians Diabetes dataset (Smith & Dickson, 1988).
Dataset Splits No The paper describes the use of synthetic and real-world datasets (Lung Cancer Dataset, Pima Indians Diabetes dataset) but does not provide specific details on how these datasets were split into training, validation, or testing sets.
Hardware Specification No The paper does not provide any specific details regarding the hardware used for conducting the experiments, such as GPU models, CPU types, or cloud computing resources.
Software Dependencies No The paper does not provide specific software dependency versions (e.g., programming language versions, library versions, or specific solver versions) used in the experiments.
Experiment Setup Yes The algorithm converges around 200 iterations. To approximate the CCE, we iteratively search with 100 values of β0 [0, 1] and β1 [0, 0.5]. The result is shown in Figure 8. Then we take the CCE as the minimum entropy H(W) such that both I(Y ; Z|X, W) and I(Z; W) are smaller than the threshold 1e 5.