Configuration files

Basic usage

Inference objects can be directly created from configuration files which facilitates the reproducibility and consistency of analyses, and simplifies the management of hyperparameters and other settings:

from fastdfe import BaseInference, Config

config = Config.from_file('config.yaml')

inf = BaseInference.from_config(config)

A configuration file typically looks like this, and reflects the parameters of BaseInference, JointInference and Config.

# The DFE parametrization to be used.
# GammaExpParametrization is parametrized by S_b, S_d, p_b, and b
model: GammaExpParametrization
# The parameters held fixed. Note that we need to
# specify a type, even when only one type is present.
# For BaseInference we thus resort to the 'all' type
fixed_params:
  all:
    eps: 0
# The initial values. This is not very important
# as fastDFE is fast and usually runs the numerical
# optimization multiple times for different initial values
x0:
  all:
    S_b: 2
    S_d: -10000
    p_b: 0.1
    b: 0.2
    eps: 0
# The number of independent runs
n_runs: 10
# The parameter bounds. Note that we do not specify the
# type here. This property should be left untouched in
# most cases.
bounds: { }
# The scales over which the parameters are optimized,
# i.e. 'lin', 'log', or 'symlog'.
scales: { }
# Whether to perform bootstraps directly when invoking
# Inference.run()
do_bootstrap: true
# The number of bootstraps
n_bootstraps: 100
# Whether to parallelize the computations across all
# available cores
parallelize: true
# The seed for the random number generator which is
# used for sampling the bootstraps and changing the
# initial values among other this. Fixing the seed
# guarantees deterministic behaviour
seed: 0
# The neutral SFS for the different types. Here we just
# have the 'all' type.
sfs_neut:
  all:
    - 173705
    - 3294
    - 1115
    - 534
    - 326
    - 239
    - 225
    - 214
    - 231
    - 176
    - 73
# The selected SFS for the different types.
sfs_sel:
  all:
    - 792884
    - 4762
    - 1397
    - 720
    - 423
    - 259
    - 247
    - 256
    - 265
    - 209
    - 90

For all available configuration options, see the Config class.

You can also create a config file from an already existing inference object:

inf.create_config().to_file('config.yaml')

JSON files

You can also use JSON by calling from_json() which I personally find more readable. A JSON configuration file would typically like this:

{
  "model": "GammaExpParametrization",
  "fixed_params": {
    "all": {
      "eps": 0
    }
  },
  "x0": {
    "all": {
      "S_b": 2,
      "S_d": -10000,
      "p_b": 0.1,
      "b": 0.2,
      "eps": 0
    }
  },
  "n_runs": 10,
  "bounds": {},
  "scales": {},
  "do_bootstrap": true,
  "n_bootstraps": 100,
  "parallelize": true,
  "seed": 0,
  "sfs_neut": {
    "all": [
      173705,
      3294,
      1115,
      534,
      326,
      239,
      225,
      214,
      231,
      176,
      73
    ]
  },
  "sfs_sel": {
    "all": [
      792884,
      4762,
      1397,
      720,
      423,
      259,
      247,
      256,
      265,
      209,
      90
    ]
  }
}

Joint inference example

A more involved configuration files configuring a joint inference with a number of fixed and shared parameters as well as covariates might look like this:

model: GammaExpParametrization
# parameter b is shared among all types
shared_params:
  - types: all
    params:
      - b
fixed_params:
  # eps is fixed for all types
  all:
    eps: 0
  # S_b is fixed for pubescens
  pubescens:
    S_b: 90
# parameter S_d has covariates
covariates:
  - param: S_d
    values:
      pendula: -32623.595481483513
      pubescens: -426.59080558648185
bounds: { }
scales: { }
x0: { }
do_bootstrap: true
linearized: true
loss_type: likelihood
n_bootstraps: 100
n_runs: 10
opts_mle: { }
parallelize: true
seed: 0
sfs_neut:
  pendula:
    - 177130
    - 997
    - 441
    - 228
    - 156
    - 117
    - 114
    - 83
    - 105
    - 109
    - 652
  pubescens:
    - 172528
    - 3612
    - 1359
    - 790
    - 584
    - 427
    - 325
    - 234
    - 166
    - 76
    - 31
sfs_sel:
  pendula:
    - 797939
    - 1329
    - 499
    - 265
    - 162
    - 104
    - 117
    - 90
    - 94
    - 119
    - 794
  pubescens:
    - 791106
    - 5326
    - 1741
    - 1005
    - 756
    - 546
    - 416
    - 294
    - 177
    - 104
    - 41