Possible convergence issue with dyadnoise

[adamhaber]

I'm working with noisy network data, and I'm interested in explicitly modeling the ties measurement error. I have reasonable prior knowledge regarding the error rates (estimating a non-tie as a tie or vice versa).

I was quite excited when I saw dyadnoise - I think this is exactly what I am looking for. Before trying this functionality on my network, I wanted to play with the various "standard" datasets.

I ran the following code:

data(faux.mesa.high) 
mesa <- faux.mesa.high
fauxmodel.01 <- ergm(mesa ~edges + 
                       nodefactor('Grade') + nodematch('Grade',diff=T) +
                       nodefactor('Race') + nodematch('Race',diff=T))
fauxmodel.02 <- ergm(mesa ~edges + 
                       nodefactor('Grade') + nodematch('Grade',diff=T) +
                       nodefactor('Race') + nodematch('Race',diff=T), obs.constraints = ~dyadnoise(0.01, 0.01)) 

and got:

Best valid proposal ‘dyadnoiseTNT’ cannot take into account hint(s) ‘sparse’.
Observed statistic(s) nodematch.Race.Black and nodematch.Race.Other are at their smallest attainable values. Their coefficients will be fixed at -Inf.
Starting contrastive divergence estimation via CD-MCMLE:
Number of informative dyads is too low. Using default imputation density.
Iteration 1 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 0.1873.
Iteration 2 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 0.4424.
Iteration 3 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 0.8789.
Iteration 4 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 0.8108.
Iteration 5 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 0.8777.
Iteration 6 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.197.
Iteration 7 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.426.
Iteration 8 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.485.
Iteration 9 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.157.
Iteration 10 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.145.
Iteration 11 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.248.
Iteration 12 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.416.
Iteration 13 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.398.
Iteration 14 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.564.
Iteration 15 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.623.
Iteration 16 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.798.
Iteration 17 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.739.
Iteration 18 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.305.
Iteration 19 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.69.
Iteration 20 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.388.
Iteration 21 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.393.
Iteration 22 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.427.
Iteration 23 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.303.
Iteration 24 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.76.
Iteration 25 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.437.
Iteration 26 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.379.
Iteration 27 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.375.
Iteration 28 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.437.
Iteration 29 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.759.
Iteration 30 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.51.
Iteration 31 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.782.
Iteration 32 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.567.
Iteration 33 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.432.
Iteration 34 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.394.
Iteration 35 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.37.
Iteration 36 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.436.
Iteration 37 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.719.
Iteration 38 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.345.
Iteration 39 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.5.
Iteration 40 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.803.
Iteration 41 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.476.
Iteration 42 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.38.
Iteration 43 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.411.
Iteration 44 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.914.
Iteration 45 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.538.
Iteration 46 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.407.
Iteration 47 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.738.
Iteration 48 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.507.
Iteration 49 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.488.
Iteration 50 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.447.
Iteration 51 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.384.
Iteration 52 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.455.
Iteration 53 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.758.
Iteration 54 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.424.
Iteration 55 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.678.
Iteration 56 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.42.
Iteration 57 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.443.
Iteration 58 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.373.
Iteration 59 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.373.
Iteration 60 of at most 60:
Convergence test P-value:0e+00
The log-likelihood improved by 1.376.
Finished CD.
Starting Monte Carlo maximum likelihood estimation (MCMLE):
Number of informative dyads is too low. Using default imputation density.
Iteration 1 of at most 60:
Model statistics ‘nodefactor.Grade.8’, ‘nodefactor.Grade.9’, ‘nodefactor.Grade.12’, ‘nodematch.Grade.7’, ‘nodematch.Grade.8’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, ‘nodematch.Grade.12’, ‘nodematch.Race.Hisp’, and ‘nodematch.Race.White’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Grade.9’, ‘nodematch.Grade.7’, and ‘nodematch.Grade.8’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.0301.
The log-likelihood improved by 0.0138.
Estimating equations are not within tolerance region.
Iteration 2 of at most 60:
Model statistics ‘nodefactor.Grade.8’, ‘nodefactor.Grade.9’, ‘nodefactor.Grade.12’, ‘nodematch.Grade.7’, ‘nodematch.Grade.8’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, ‘nodematch.Grade.12’, ‘nodematch.Race.Hisp’, and ‘nodematch.Race.White’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Grade.9’, ‘nodematch.Grade.7’, and ‘nodematch.Grade.8’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.3604.
The log-likelihood improved by 0.9301.
Estimating equations are not within tolerance region.
Iteration 3 of at most 60:
Model statistics ‘nodefactor.Grade.8’, ‘nodematch.Grade.7’, ‘nodematch.Grade.8’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, ‘nodematch.Grade.12’, ‘nodematch.Race.Hisp’, and ‘nodematch.Race.White’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Grade.11’, ‘nodematch.Grade.9’, and ‘nodematch.Grade.10’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.1040.
The log-likelihood improved by 1.0065.
Estimating equations are not within tolerance region.
Iteration 4 of at most 60:
Model statistics ‘nodefactor.Grade.8’, ‘nodefactor.Grade.12’, ‘nodematch.Grade.7’, ‘nodematch.Grade.8’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, ‘nodematch.Grade.12’, ‘nodematch.Race.Hisp’, and ‘nodematch.Race.White’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Grade.10’, ‘nodematch.Grade.8’, and ‘nodematch.Grade.9’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.1213.
The log-likelihood improved by 1.6849.
Estimating equations are not within tolerance region.
Estimating equations did not move closer to tolerance region more than 1 time(s) in 4 steps; increasing sample size.
Iteration 5 of at most 60:
Model statistics ‘nodematch.Grade.7’, ‘nodematch.Grade.8’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, ‘nodematch.Grade.12’, ‘nodematch.Race.Hisp’, and ‘nodematch.Race.White’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Grade.12’ and ‘nodematch.Grade.11’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.0248.
The log-likelihood improved by 0.0716.
Estimating equations are not within tolerance region.
Iteration 6 of at most 60:
Model statistics ‘nodefactor.Grade.12’, ‘nodematch.Grade.7’, ‘nodematch.Grade.8’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, ‘nodematch.Grade.12’, ‘nodematch.Race.Hisp’, and ‘nodematch.Race.White’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Grade.11’ and ‘nodematch.Grade.10’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.0079.
The log-likelihood improved by 0.0084.
Estimating equations are not within tolerance region.
Estimating equations did not move closer to tolerance region more than 1 time(s) in 4 steps; increasing sample size.
Iteration 7 of at most 60:
Model statistics ‘nodefactor.Grade.8’, ‘nodematch.Grade.7’, ‘nodematch.Grade.8’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, ‘nodematch.Grade.12’, and ‘nodematch.Race.White’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Grade.11’, ‘nodematch.Grade.9’, and ‘nodematch.Grade.11’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.1489.
The log-likelihood improved by 2.5989.
Estimating equations are not within tolerance region.
Iteration 8 of at most 60:
Model statistics ‘nodematch.Grade.7’, ‘nodematch.Grade.8’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, ‘nodematch.Grade.12’, and ‘nodematch.Race.White’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Grade.12’, ‘nodematch.Grade.10’, and ‘nodematch.Grade.12’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.1191.
The log-likelihood improved by 2.2449.
Estimating equations are not within tolerance region.
Iteration 9 of at most 60:
Model statistics ‘nodematch.Grade.7’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, and ‘nodematch.Grade.12’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Race.Hisp’ and ‘nodefactor.Race.NatAm’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.0797.
The log-likelihood improved by 1.5492.
Estimating equations are not within tolerance region.
Estimating equations did not move closer to tolerance region more than 1 time(s) in 4 steps; increasing sample size.
Iteration 10 of at most 60:
Model statistics ‘nodematch.Grade.7’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, and ‘nodematch.Grade.12’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Race.Hisp’ and ‘nodefactor.Race.NatAm’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.1003.
The log-likelihood improved by 2.4228.
Estimating equations are not within tolerance region.
Iteration 11 of at most 60:
Model statistics ‘nodematch.Grade.7’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, and ‘nodematch.Grade.12’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Race.NatAm’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.1055.
The log-likelihood improved by 2.5077.
Estimating equations are not within tolerance region.
Estimating equations did not move closer to tolerance region more than 1 time(s) in 4 steps; increasing sample size.
Iteration 12 of at most 60:
Model statistics ‘nodematch.Grade.7’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, and ‘nodematch.Grade.12’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodematch.Grade.7’, ‘nodefactor.Race.Hisp’, and ‘nodefactor.Race.NatAm’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.1068.
The log-likelihood improved by 2.4985.
Estimating equations are not within tolerance region.
Iteration 13 of at most 60:
Model statistics ‘nodematch.Grade.7’, ‘nodematch.Grade.9’, ‘nodematch.Grade.11’, and ‘nodematch.Grade.12’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodematch.Grade.8’ and ‘nodefactor.Race.Other’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.1222.
The log-likelihood improved by 3.0557.
Estimating equations are not within tolerance region.
Estimating equations did not move closer to tolerance region more than 1 time(s) in 4 steps; increasing sample size.
Iteration 14 of at most 60:
Model statistics ‘nodematch.Grade.7’, ‘nodematch.Grade.9’, ‘nodematch.Grade.11’, and ‘nodematch.Grade.12’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Race.Other’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.1268.
The log-likelihood improved by 3.2603.
Estimating equations are not within tolerance region.
Iteration 15 of at most 60:
Model statistics ‘nodematch.Grade.7’, ‘nodematch.Grade.9’, ‘nodematch.Grade.10’, ‘nodematch.Grade.11’, and ‘nodematch.Grade.12’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Race.NatAm’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.1126.
The log-likelihood improved by 2.6938.
Estimating equations are not within tolerance region.
Estimating equations did not move closer to tolerance region more than 1 time(s) in 4 steps; increasing sample size.
Iteration 16 of at most 60:
Model statistics ‘nodematch.Grade.7’, ‘nodematch.Grade.9’, ‘nodematch.Grade.11’, and ‘nodematch.Grade.12’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodematch.Grade.8’ and ‘nodefactor.Race.Other’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.1087.
The log-likelihood improved by 3.3860.
Estimating equations are not within tolerance region.
Iteration 17 of at most 60:
Model statistics ‘nodematch.Grade.7’, ‘nodematch.Grade.9’, ‘nodematch.Grade.11’, and ‘nodematch.Grade.12’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodefactor.Race.Other’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length 0.0776.
The log-likelihood improved by 2.4542.
Estimating equations are not within tolerance region.
Estimating equations did not move closer to tolerance region more than 1 time(s) in 4 steps; increasing sample size.
Iteration 18 of at most 60:
Model statistics ‘nodematch.Grade.7’, ‘nodematch.Grade.9’, ‘nodematch.Grade.11’, and ‘nodematch.Grade.12’ are not varying. This may indicate that the observed data occupies an extreme point in the sample space or that the estimation has reached a dead-end configuration.
Warning: Model statistics ‘nodematch.Grade.8’, ‘nodefactor.Race.NatAm’, and ‘nodefactor.Race.Other’ are linear combinations of some set of preceding statistics at the current stage of the estimation. This may indicate that the model is nonidentifiable.
Optimizing with step length < 0.0001.
Error in h(simpleError(msg, call)) : 
  error in evaluating the argument 'object' in selecting a method for function 'isSymmetric': non-conformable arrays

is this the expected behavior?

[krivit]

So, at first glance, I had thought that the issue was a matter of figuring out how to specify this correctly, so I moved this to a discussion, but it looks like there are some bugs that this reveals. In particular, we probably need to add a unit test that forces an observation process in a situation with 0 missing dyads, since that's the case that this is dealing with.

In the meantime, @adamhaber , does the fitting work for simple models (warnings notwithstanding)?