Prior State Distributions#
Given samples drawn during an MCMC round, compute the (prior) state distributions.
This is done for each sample and for a list of specified scenarios. The computation is
cached at a location specified by the --cache_dir argument using joblib.
- lyscripts.compute.priors.compute_priors(model_config: ModelConfig, graph_config: GraphConfig, dist_configs: dict[str, DistributionConfig], samples: ndarray, t_stages: list[int | str], t_stages_dist: list[float], mode: Literal['HMM', 'BN'] = 'HMM', progress_desc: str = 'Computing priors from samples') ndarray[source]#
Compute prior state distributions from the
samplesfor themodel.This will call the
modelmethodstate_dist()for each of thesamples. The prior state distributions are computed for each of thet_stagesand marginalized over using thet_stages_dist.
- pydantic settings lyscripts.compute.priors.PriorsCLI[source]#
Bases:
BaseComputeCLICompute the prior state distributions from MCMC samples.
Show JSON schema
{ "title": "PriorsCLI", "description": "Compute the prior state distributions from MCMC samples.", "type": "object", "properties": { "configs": { "default": [ "config.yaml" ], "description": "Path to the YAML file(s) that contain the configuration(s). Configs from YAML files may be overwritten by command line arguments. When multiple files are specified, the configs are merged in the order they are given. Note that every config file must have a `version: 1` key in it.", "items": { "format": "path", "type": "string" }, "title": "Configs", "type": "array" }, "graph": { "$ref": "#/$defs/GraphConfig" }, "model": { "$ref": "#/$defs/ModelConfig", "default": { "external_file": null, "class_name": "Unilateral", "constructor": "binary", "max_time": 10, "named_params": null, "kwargs": {} } }, "distributions": { "additionalProperties": { "$ref": "#/$defs/DistributionConfig" }, "default": {}, "description": "Mapping of model T-categories to predefined distributions over diagnose times.", "title": "Distributions", "type": "object" }, "cache_dir": { "default": "/home/docs/checkouts/readthedocs.org/user_builds/lyscripts/checkouts/latest/docs/source/.cache", "description": "Cache directory for storing function calls.", "format": "path", "title": "Cache Dir", "type": "string" }, "scenarios": { "default": [], "description": "List of scenarios to compute risks for.", "items": { "$ref": "#/$defs/ScenarioConfig" }, "title": "Scenarios", "type": "array" }, "sampling": { "$ref": "#/$defs/SamplingConfig" }, "priors": { "$ref": "#/$defs/HDF5FileStorage", "description": "Storage for the computed priors." } }, "$defs": { "DiagnosisConfig": { "description": "Defines an ipsi- and contralateral diagnosis pattern.", "properties": { "ipsi": { "additionalProperties": { "additionalProperties": { "anyOf": [ { "enum": [ false, 0, "healthy", true, 1, "involved", "micro", "macro", "notmacro" ] }, { "type": "null" } ] }, "type": "object" }, "default": {}, "description": "Observed diagnoses by different modalities on the ipsi neck.", "examples": [ { "CT": { "II": true, "III": false } } ], "title": "Ipsi", "type": "object" }, "contra": { "additionalProperties": { "additionalProperties": { "anyOf": [ { "enum": [ false, 0, "healthy", true, 1, "involved", "micro", "macro", "notmacro" ] }, { "type": "null" } ] }, "type": "object" }, "default": {}, "description": "Observed diagnoses by different modalities on the contra neck.", "title": "Contra", "type": "object" } }, "title": "DiagnosisConfig", "type": "object" }, "DistributionConfig": { "description": "Configuration defining a distribution over diagnose times.", "properties": { "kind": { "default": "frozen", "description": "Parametric distributions may be updated.", "enum": [ "frozen", "parametric" ], "title": "Kind", "type": "string" }, "func": { "const": "binomial", "default": "binomial", "description": "Name of predefined function to use as distribution.", "title": "Func", "type": "string" }, "params": { "additionalProperties": { "anyOf": [ { "type": "integer" }, { "type": "number" } ] }, "default": {}, "description": "Parameters to pass to the predefined function.", "title": "Params", "type": "object" } }, "title": "DistributionConfig", "type": "object" }, "GraphConfig": { "description": "Specifies how the tumor(s) and LNLs are connected in a DAG.", "properties": { "tumor": { "additionalProperties": { "items": { "type": "string" }, "type": "array" }, "description": "Define the name of the tumor(s) and which LNLs it/they drain to.", "title": "Tumor", "type": "object" }, "lnl": { "additionalProperties": { "items": { "type": "string" }, "type": "array" }, "description": "Define the name of the LNL(s) and which LNLs it/they drain to.", "title": "Lnl", "type": "object" } }, "required": [ "tumor", "lnl" ], "title": "GraphConfig", "type": "object" }, "HDF5FileStorage": { "description": "HDF5 file storage for in- and outputs of computations.", "properties": { "file": { "description": "Path to the HDF5 file. Parent directories are created if needed.", "format": "path", "title": "File", "type": "string" }, "dataset": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "description": "Name of the dataset in the HDF5 file. Save/load methods can override this.", "title": "Dataset" } }, "required": [ "file" ], "title": "HDF5FileStorage", "type": "object" }, "InvolvementConfig": { "description": "Config that defines an ipsi- and contralateral involvement pattern.", "properties": { "ipsi": { "additionalProperties": { "anyOf": [ { "enum": [ false, 0, "healthy", true, 1, "involved", "micro", "macro", "notmacro" ] }, { "type": "null" } ] }, "default": {}, "description": "Involvement pattern for the ipsilateral side of the neck.", "examples": [ { "II": true, "III": false } ], "title": "Ipsi", "type": "object" }, "contra": { "additionalProperties": { "anyOf": [ { "enum": [ false, 0, "healthy", true, 1, "involved", "micro", "macro", "notmacro" ] }, { "type": "null" } ] }, "default": {}, "description": "Involvement pattern for the contralateral side of the neck.", "title": "Contra", "type": "object" } }, "title": "InvolvementConfig", "type": "object" }, "ModelConfig": { "description": "Define which of the ``lymph`` models to use and how to set them up.", "properties": { "external_file": { "anyOf": [ { "format": "file-path", "type": "string" }, { "type": "null" } ], "default": null, "description": "Path to a Python file that defines a model.", "title": "External File" }, "class_name": { "default": "Unilateral", "description": "Name of the model class to use.", "enum": [ "Unilateral", "Bilateral", "Midline" ], "title": "Class Name", "type": "string" }, "constructor": { "default": "binary", "description": "Trinary models differentiate btw. micro- and macroscopic disease.", "enum": [ "binary", "trinary" ], "title": "Constructor", "type": "string" }, "max_time": { "default": 10, "description": "Max. number of time-steps to evolve the model over.", "title": "Max Time", "type": "integer" }, "named_params": { "default": null, "description": "Subset of valid model parameters a sampler may provide in the form of a dictionary to the model instead of as an array. Or, after sampling, with this list, one may safely recover which parameter corresponds to which index in the sample.", "items": { "type": "string" }, "title": "Named Params", "type": "array" }, "kwargs": { "additionalProperties": true, "default": {}, "description": "Additional keyword arguments to pass to the model constructor.", "title": "Kwargs", "type": "object" } }, "title": "ModelConfig", "type": "object" }, "SamplingConfig": { "description": "Settings to configure the MCMC sampling.", "properties": { "storage_file": { "description": "Path to HDF5 file store results or load last state.", "format": "path", "title": "Storage File", "type": "string" }, "history_file": { "anyOf": [ { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "description": "Path to store the burn-in metrics (as CSV file).", "title": "History File" }, "dataset": { "default": "mcmc", "description": "Name of the dataset in the HDF5 file.", "title": "Dataset", "type": "string" }, "cores": { "anyOf": [ { "exclusiveMinimum": 0, "type": "integer" }, { "type": "null" } ], "default": 2, "description": "Number of cores to use for parallel sampling. If `None`, no parallel processing is used.", "title": "Cores" }, "seed": { "default": 42, "description": "Seed for the random number generator.", "title": "Seed", "type": "integer" }, "walkers_per_dim": { "default": 20, "description": "Number of walkers per parameter space dimension.", "title": "Walkers Per Dim", "type": "integer" }, "check_interval": { "default": 50, "description": "Check for convergence each time after this many steps.", "title": "Check Interval", "type": "integer" }, "trust_factor": { "default": 50.0, "description": "Trust the autocorrelation time only when it's smaller than this factor times the length of the chain.", "title": "Trust Factor", "type": "number" }, "relative_thresh": { "default": 0.05, "description": "Relative threshold for convergence.", "title": "Relative Thresh", "type": "number" }, "burnin_steps": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "description": "Number of burn-in steps to take. If None, burn-in runs until convergence.", "title": "Burnin Steps" }, "num_steps": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": 100, "description": "Number of steps to take in the MCMC sampling.", "title": "Num Steps" }, "thin_by": { "default": 10, "description": "How many samples to draw before for saving one.", "title": "Thin By", "type": "integer" }, "inverse_temp": { "default": 1.0, "description": "Inverse temperature for thermodynamic integration. Note that this is not yet fully implemented.", "title": "Inverse Temp", "type": "number" } }, "required": [ "storage_file" ], "title": "SamplingConfig", "type": "object" }, "ScenarioConfig": { "description": "Define a scenario for which e.g. prevalences and risks may be computed.", "properties": { "t_stages": { "description": "List of T-stages to marginalize over in the scenario.", "examples": [ [ "early" ], [ 3, 4 ] ], "items": { "anyOf": [ { "type": "integer" }, { "type": "string" } ] }, "title": "T Stages", "type": "array" }, "t_stages_dist": { "default": [ 1.0 ], "description": "Distribution over T-stages to use for marginalization.", "examples": [ [ 1.0 ], [ 0.6, 0.4 ] ], "items": { "type": "number" }, "title": "T Stages Dist", "type": "array" }, "midext": { "anyOf": [ { "type": "boolean" }, { "type": "null" } ], "default": null, "description": "Whether the patient's tumor extends over the midline.", "title": "Midext" }, "mode": { "default": "HMM", "description": "Which underlying model architecture to use.", "enum": [ "HMM", "BN" ], "title": "Mode", "type": "string" }, "involvement": { "$ref": "#/$defs/InvolvementConfig", "default": { "ipsi": {}, "contra": {} } }, "diagnosis": { "$ref": "#/$defs/DiagnosisConfig", "default": { "ipsi": {}, "contra": {} } } }, "required": [ "t_stages" ], "title": "ScenarioConfig", "type": "object" } }, "required": [ "graph", "sampling", "priors" ] }
- field priors: HDF5FileStorage [Required]#
Storage for the computed priors.
- cli_cmd() None[source]#
Start the
priorssubcommand.Given a
graph,model,distributionsover diagnosis times, and MCMC samples loaded from thesamplingargument, this command computes the prior state distributions for each of the specifiedscenarios.Precomputing these state distributions is useful, because they largely only depend on T-stage and not on the diagnosis or involvement of interest. Hence, computing the
posteriorsandriskscan be sped up.Note that this command will use joblib to cache its computations.
Command Help#
Usage: lyscripts compute priors [-h] [--configs list[Path]] [--graph [JSON]]
[--graph.tumor dict[str,list[str]]]
[--graph.lnl dict[str,list[str]]]
[--model [JSON]]
[--model.external-file {Path,null}]
[--model.class-name {Unilateral,Bilateral,Midline}]
[--model.constructor {binary,trinary}]
[--model.max-time int]
[--model.named-params Sequence[str]]
[--model.kwargs dict[str,Any]]
[--distributions dict[str,JSON]]
[--cache-dir Path] [--scenarios list[JSON]]
[--sampling [JSON]]
[--sampling.storage-file Path]
[--sampling.history-file {Path,null}]
[--sampling.dataset str]
[--sampling.cores {int,null}]
[--sampling.seed int]
[--sampling.walkers-per-dim int]
[--sampling.check-interval int]
[--sampling.trust-factor float]
[--sampling.relative-thresh float]
[--sampling.burnin-steps {int,null}]
[--sampling.num-steps {int,null}]
[--sampling.thin-by int]
[--sampling.inverse-temp float]
[--priors [JSON]] [--priors.file Path]
[--priors.dataset {str,null}]
Compute the prior state distributions from MCMC samples.
Options:
-h, --help show this help message and exit
--configs list[Path] Path to the YAML file(s) that contain the
configuration(s). Configs from YAML files may be
overwritten by command line arguments. When multiple
files are specified, the configs are merged in the
order they are given. Note that every config file must
have a `version: 1` key in it. (default:
['config.yaml'])
--distributions dict[str,JSON]
Mapping of model T-categories to predefined
distributions over diagnose times. (default: {})
--cache-dir Path Cache directory for storing function calls. (default:
/home/docs/checkouts/readthedocs.org/user_builds/lyscr
ipts/checkouts/latest/docs/source/.cache)
--scenarios list[JSON]
List of scenarios to compute risks for. (default: [])
Graph Options:
Specifies how the tumor(s) and LNLs are connected in a DAG.
--graph [JSON] set graph from JSON string (default: {})
--graph.tumor dict[str,list[str]]
Define the name of the tumor(s) and which LNLs it/they
drain to. (required)
--graph.lnl dict[str,list[str]]
Define the name of the LNL(s) and which LNLs it/they
drain to. (required)
Model Options:
Define which of the ``lymph`` models to use and how to set them up.
--model [JSON] set model from JSON string (default: {})
--model.external-file {Path,null}
Path to a Python file that defines a model. (default:
None)
--model.class-name {Unilateral,Bilateral,Midline}
Name of the model class to use. (default: Unilateral)
--model.constructor {binary,trinary}
Trinary models differentiate btw. micro- and
macroscopic disease. (default: binary)
--model.max-time int Max. number of time-steps to evolve the model over.
(default: 10)
--model.named-params Sequence[str]
Subset of valid model parameters a sampler may provide
in the form of a dictionary to the model instead of as
an array. Or, after sampling, with this list, one may
safely recover which parameter corresponds to which
index in the sample. (default: None)
--model.kwargs dict[str,Any]
Additional keyword arguments to pass to the model
constructor. (default: {})
Sampling Options:
Settings to configure the MCMC sampling.
--sampling [JSON] set sampling from JSON string (default: {})
--sampling.storage-file Path
Path to HDF5 file store results or load last state.
(required)
--sampling.history-file {Path,null}
Path to store the burn-in metrics (as CSV file).
(default: null)
--sampling.dataset str
Name of the dataset in the HDF5 file. (default: mcmc)
--sampling.cores {int,null}
Number of cores to use for parallel sampling. If
`None`, no parallel processing is used. (default: 2)
--sampling.seed int Seed for the random number generator. (default: 42)
--sampling.walkers-per-dim int
Number of walkers per parameter space dimension.
(default: 20)
--sampling.check-interval int
Check for convergence each time after this many steps.
(default: 50)
--sampling.trust-factor float
Trust the autocorrelation time only when it's smaller
than this factor times the length of the chain.
(default: 50.0)
--sampling.relative-thresh float
Relative threshold for convergence. (default: 0.05)
--sampling.burnin-steps {int,null}
Number of burn-in steps to take. If None, burn-in runs
until convergence. (default: null)
--sampling.num-steps {int,null}
Number of steps to take in the MCMC sampling.
(default: 100)
--sampling.thin-by int
How many samples to draw before for saving one.
(default: 10)
--sampling.inverse-temp float
Inverse temperature for thermodynamic integration.
Note that this is not yet fully implemented. (default:
1.0)
Priors Options:
HDF5 file storage for in- and outputs of computations.
--priors [JSON] set priors from JSON string (default: {})
--priors.file Path Path to the HDF5 file. Parent directories are created
if needed. (required)
--priors.dataset {str,null}
Name of the dataset in the HDF5 file. Save/load
methods can override this. (default: null)