Releases: rmnldwg/lynference
win-graph-v1
Previously, we computed the BIC for unilateral models differing in their underlying graph definition. Among the best performing graphs was the one tested in extended-add21-add35-v1. For the base graph, tested in extended-base-v1 and this "winning graph", we are now doing an extended analysis on the basis of the log-evidence, computed using an expensive thermodynamic integration with 64 steps.
The investigated graph in this release is shown below:
The results of the comparison are tabulated below:
| graph | log-evidence | max. llh | |
|---|---|---|---|
base-graph-v1 |
|||
win-graph-v1 (this) |
base-graph-v1
Previously, we computed the BIC for unilateral models differing in their underlying graph definition. Among the best performing graphs was the one tested in extended-add21-add35-v1. For the base graph, tested in extended-base-v1 and this "winning graph", we are now doing an extended analysis on the basis of the log-evidence, computed using an expensive thermodynamic integration with 64 steps.
The investigated graph in this release is shown below:
The results of the comparison are tabulated below:
| graph | log-evidence | max. llh | |
|---|---|---|---|
base-graph-v1 (this) |
|||
win-graph-v1 |
extended-add21-add45-v1
New data allowed us to include the LNLs V and VII into the model. Consequently, we ran some experiments to compare different graphs, connecting the tumor and the LNLs among each other.
To be able to compare them, we computed the BIC, the maximum likelihood, and the mean likelihood. Using thermodynamic integration, we reassured ourselves that - at least for the extended-base-v1 model - the negative log-evidence and the BIC/2 agreed well.
The investigated graph is show below:
Graph: Network of tumor and lymph node levels. Red arcs depict spread from tumor to the individual LNLs, which were considered in all models. Blue arcs represent spread probabilities from LNL to LNL that were also present in all investigated graphs. The green edges, however, were the ones we added one by one to compare the resulting model's performance.
Results
From the model comparison values tabulated below, we inferred that the model with additional arcs from LNL II to LNL I and from LNL III to V (extended-add21-add35-v1) was the best one for our purposes.
| Model | - log-evidence | BIC/2 | - max. llh | - mean llh |
|---|---|---|---|---|
| extended-base-v1 | 1212.3 ± 3.7 | 1212.0 | 1180.4 | 1184.7 |
| extended-add12-v1 | 1211.7 | 1177.0 | 1180.8 | |
| extended-add21-v1 | 1208.2 | 1173.5 | 1177.9 | |
| extended-add35-v1 | 1212.4 | 1177.7 | 1182.1 | |
| extended-add45-v1 | 1213.6 | 1178.9 | 1183.1 | |
| extended-add27-v1 | 1214.9 | 1180.2 | 1184.4 | |
| extended-add21-add35-v1 | 1208.4 | 1170.5 | 1174.7 | |
| extended-add21-add45-v1 | 1209.7 | 1171.8 | 1176.0 | |
| extended-add21-add27-v1 | 1210.7 | 1172.8 | 1177.5 |
extended-add21-add35-v1
New data allowed us to include the LNLs V and VII into the model. Consequently, we ran some experiments to compare different graphs, connecting the tumor and the LNLs among each other.
To be able to compare them, we computed the BIC, the maximum likelihood, and the mean likelihood. Using thermodynamic integration, we reassured ourselves that - at least for the extended-base-v1 model - the negative log-evidence and the BIC/2 agreed well.
The investigated graph is show below:
Graph: Network of tumor and lymph node levels. Red arcs depict spread from tumor to the individual LNLs, which were considered in all models. Blue arcs represent spread probabilities from LNL to LNL that were also present in all investigated graphs. The green edges, however, were the ones we added one by one to compare the resulting model's performance.
Results
From the model comparison values tabulated below, we inferred that the model with additional arcs from LNL II to LNL I and from LNL III to V (extended-add21-add35-v1) was the best one for our purposes.
| Model | - log-evidence | BIC/2 | - max. llh | - mean llh |
|---|---|---|---|---|
| extended-base-v1 | 1212.3 ± 3.7 | 1212.0 | 1180.4 | 1184.7 |
| extended-add12-v1 | 1211.7 | 1177.0 | 1180.8 | |
| extended-add21-v1 | 1208.2 | 1173.5 | 1177.9 | |
| extended-add35-v1 | 1212.4 | 1177.7 | 1182.1 | |
| extended-add45-v1 | 1213.6 | 1178.9 | 1183.1 | |
| extended-add27-v1 | 1214.9 | 1180.2 | 1184.4 | |
| extended-add21-add35-v1 | 1208.4 | 1170.5 | 1174.7 | |
| extended-add21-add45-v1 | 1209.7 | 1171.8 | 1176.0 | |
| extended-add21-add27-v1 | 1210.7 | 1172.8 | 1177.5 |
extended-add21-add27-v1
New data allowed us to include the LNLs V and VII into the model. Consequently, we ran some experiments to compare different graphs, connecting the tumor and the LNLs among each other.
To be able to compare them, we computed the BIC, the maximum likelihood, and the mean likelihood. Using thermodynamic integration, we reassured ourselves that - at least for the extended-base-v1 model - the negative log-evidence and the BIC/2 agreed well.
The investigated graph is show below:
Graph: Network of tumor and lymph node levels. Red arcs depict spread from tumor to the individual LNLs, which were considered in all models. Blue arcs represent spread probabilities from LNL to LNL that were also present in all investigated graphs. The green edges, however, were the ones we added one by one to compare the resulting model's performance.
Results
From the model comparison values tabulated below, we inferred that the model with additional arcs from LNL II to LNL I and from LNL III to V (extended-add21-add35-v1) was the best one for our purposes.
| Model | - log-evidence | BIC/2 | - max. llh | - mean llh |
|---|---|---|---|---|
| extended-base-v1 | 1212.3 ± 3.7 | 1212.0 | 1180.4 | 1184.7 |
| extended-add12-v1 | 1211.7 | 1177.0 | 1180.8 | |
| extended-add21-v1 | 1208.2 | 1173.5 | 1177.9 | |
| extended-add35-v1 | 1212.4 | 1177.7 | 1182.1 | |
| extended-add45-v1 | 1213.6 | 1178.9 | 1183.1 | |
| extended-add27-v1 | 1214.9 | 1180.2 | 1184.4 | |
| extended-add21-add35-v1 | 1208.4 | 1170.5 | 1174.7 | |
| extended-add21-add45-v1 | 1209.7 | 1171.8 | 1176.0 | |
| extended-add21-add27-v1 | 1210.7 | 1172.8 | 1177.5 |
extended-add27-v1
New data allowed us to include the LNLs V and VII into the model. Consequently, we ran some experiments to compare different graphs, connecting the tumor and the LNLs among each other.
To be able to compare them, we computed the BIC, the maximum likelihood, and the mean likelihood. Using thermodynamic integration, we reassured ourselves that - at least for the extended-base-v1 model - the negative log-evidence and the BIC/2 agreed well.
The investigated graph is show below:
Graph: Network of tumor and lymph node levels. Red arcs depict spread from tumor to the individual LNLs, which were considered in all models. Blue arcs represent spread probabilities from LNL to LNL that were also present in all investigated graphs. The green edges, however, were the ones we added one by one to compare the resulting model's performance.
Results
From the model comparison values tabulated below, we inferred that the model with additional arcs from LNL II to LNL I and from LNL III to V (extended-add21-add35-v1) was the best one for our purposes.
| Model | - log-evidence | BIC/2 | - max. llh | - mean llh |
|---|---|---|---|---|
| extended-base-v1 | 1212.3 ± 3.7 | 1212.0 | 1180.4 | 1184.7 |
| extended-add12-v1 | 1211.7 | 1177.0 | 1180.8 | |
| extended-add21-v1 | 1208.2 | 1173.5 | 1177.9 | |
| extended-add35-v1 | 1212.4 | 1177.7 | 1182.1 | |
| extended-add45-v1 | 1213.6 | 1178.9 | 1183.1 | |
| extended-add27-v1 | 1214.9 | 1180.2 | 1184.4 | |
| extended-add21-add35-v1 | 1208.4 | 1170.5 | 1174.7 | |
| extended-add21-add45-v1 | 1209.7 | 1171.8 | 1176.0 | |
| extended-add21-add27-v1 | 1210.7 | 1172.8 | 1177.5 |
extended-add45-v1
New data allowed us to include the LNLs V and VII into the model. Consequently, we ran some experiments to compare different graphs, connecting the tumor and the LNLs among each other.
To be able to compare them, we computed the BIC, the maximum likelihood, and the mean likelihood. Using thermodynamic integration, we reassured ourselves that - at least for the extended-base-v1 model - the negative log-evidence and the BIC/2 agreed well.
The investigated graph is show below:
Graph: Network of tumor and lymph node levels. Red arcs depict spread from tumor to the individual LNLs, which were considered in all models. Blue arcs represent spread probabilities from LNL to LNL that were also present in all investigated graphs. The green edges, however, were the ones we added one by one to compare the resulting model's performance.
Results
From the model comparison values tabulated below, we inferred that the model with additional arcs from LNL II to LNL I and from LNL III to V (extended-add21-add35-v1) was the best one for our purposes.
| Model | - log-evidence | BIC/2 | - max. llh | - mean llh |
|---|---|---|---|---|
| extended-base-v1 | 1212.3 ± 3.7 | 1212.0 | 1180.4 | 1184.7 |
| extended-add12-v1 | 1211.7 | 1177.0 | 1180.8 | |
| extended-add21-v1 | 1208.2 | 1173.5 | 1177.9 | |
| extended-add35-v1 | 1212.4 | 1177.7 | 1182.1 | |
| extended-add45-v1 | 1213.6 | 1178.9 | 1183.1 | |
| extended-add27-v1 | 1214.9 | 1180.2 | 1184.4 | |
| extended-add21-add35-v1 | 1208.4 | 1170.5 | 1174.7 | |
| extended-add21-add45-v1 | 1209.7 | 1171.8 | 1176.0 | |
| extended-add21-add27-v1 | 1210.7 | 1172.8 | 1177.5 |
extended-add35-v1
New data allowed us to include the LNLs V and VII into the model. Consequently, we ran some experiments to compare different graphs, connecting the tumor and the LNLs among each other.
To be able to compare them, we computed the BIC, the maximum likelihood, and the mean likelihood. Using thermodynamic integration, we reassured ourselves that - at least for the extended-base-v1 model - the negative log-evidence and the BIC/2 agreed well.
The investigated graph is show below:
Graph: Network of tumor and lymph node levels. Red arcs depict spread from tumor to the individual LNLs, which were considered in all models. Blue arcs represent spread probabilities from LNL to LNL that were also present in all investigated graphs. The green edges, however, were the ones we added one by one to compare the resulting model's performance.
Results
From the model comparison values tabulated below, we inferred that the model with additional arcs from LNL II to LNL I and from LNL III to V (extended-add21-add35-v1) was the best one for our purposes.
| Model | - log-evidence | BIC/2 | - max. llh | - mean llh |
|---|---|---|---|---|
| extended-base-v1 | 1212.3 ± 3.7 | 1212.0 | 1180.4 | 1184.7 |
| extended-add12-v1 | 1211.7 | 1177.0 | 1180.8 | |
| extended-add21-v1 | 1208.2 | 1173.5 | 1177.9 | |
| extended-add35-v1 | 1212.4 | 1177.7 | 1182.1 | |
| extended-add45-v1 | 1213.6 | 1178.9 | 1183.1 | |
| extended-add27-v1 | 1214.9 | 1180.2 | 1184.4 | |
| extended-add21-add35-v1 | 1208.4 | 1170.5 | 1174.7 | |
| extended-add21-add45-v1 | 1209.7 | 1171.8 | 1176.0 | |
| extended-add21-add27-v1 | 1210.7 | 1172.8 | 1177.5 |
extended-base-v1
New data allowed us to include the LNLs V and VII into the model. Consequently, we ran some experiments to compare different graphs, connecting the tumor and the LNLs among each other.
To be able to compare them, we computed the BIC, the maximum likelihood, and the mean likelihood. Using thermodynamic integration, we reassured ourselves that - at least for the extended-base-v1 model - the negative log-evidence and the BIC/2 agreed well.
The investigated graph is show below:
Graph: Network of tumor and lymph node levels. Red arcs depict spread from tumor to the individual LNLs, which were considered in all models. Blue arcs represent spread probabilities from LNL to LNL that were also present in all investigated graphs. The green edges, however, were the ones we added one by one to compare the resulting model's performance.
Results
From the model comparison values tabulated below, we inferred that the model with additional arcs from LNL II to LNL I and from LNL III to V (extended-add21-add35-v1) was the best one for our purposes.
| Model | - log-evidence | BIC/2 | - max. llh | - mean llh |
|---|---|---|---|---|
| extended-base-v1 | 1212.3 ± 3.7 | 1212.0 | 1180.4 | 1184.7 |
| extended-add12-v1 | 1211.7 | 1177.0 | 1180.8 | |
| extended-add21-v1 | 1208.2 | 1173.5 | 1177.9 | |
| extended-add35-v1 | 1212.4 | 1177.7 | 1182.1 | |
| extended-add45-v1 | 1213.6 | 1178.9 | 1183.1 | |
| extended-add27-v1 | 1214.9 | 1180.2 | 1184.4 | |
| extended-add21-add35-v1 | 1208.4 | 1170.5 | 1174.7 | |
| extended-add21-add45-v1 | 1209.7 | 1171.8 | 1176.0 | |
| extended-add21-add27-v1 | 1210.7 | 1172.8 | 1177.5 |
extended-add21-v1
New data allowed us to include the LNLs V and VII into the model. Consequently, we ran some experiments to compare different graphs, connecting the tumor and the LNLs among each other.
To be able to compare them, we computed the BIC, the maximum likelihood, and the mean likelihood. Using thermodynamic integration, we reassured ourselves that - at least for the extended-base-v1 model - the negative log-evidence and the BIC/2 agreed well.
The investigated graph is show below:
Graph: Network of tumor and lymph node levels. Red arcs depict spread from tumor to the individual LNLs, which were considered in all models. Blue arcs represent spread probabilities from LNL to LNL that were also present in all investigated graphs. The green edges, however, were the ones we added one by one to compare the resulting model's performance.
Results
From the model comparison values tabulated below, we inferred that the model with additional arcs from LNL II to LNL I and from LNL III to V (extended-add21-add35-v1) was the best one for our purposes.
| Model | - log-evidence | BIC/2 | - max. llh | - mean llh |
|---|---|---|---|---|
| extended-base-v1 | 1212.3 ± 3.7 | 1212.0 | 1180.4 | 1184.7 |
| extended-add12-v1 | 1211.7 | 1177.0 | 1180.8 | |
| extended-add21-v1 | 1208.2 | 1173.5 | 1177.9 | |
| extended-add35-v1 | 1212.4 | 1177.7 | 1182.1 | |
| extended-add45-v1 | 1213.6 | 1178.9 | 1183.1 | |
| extended-add27-v1 | 1214.9 | 1180.2 | 1184.4 | |
| extended-add21-add35-v1 | 1208.4 | 1170.5 | 1174.7 | |
| extended-add21-add45-v1 | 1209.7 | 1171.8 | 1176.0 | |
| extended-add21-add27-v1 | 1210.7 | 1172.8 | 1177.5 |