def generic_cccv(type, genotype, phenotype, folds, replicates, threads, comple_steps, conta_steps,
verb, groups=None, rank=None, optimize=False, out=None, n_features=None,
params_file=None,
*args, **kwargs):
"""
Perform crossvalidation over a range of simulated completeness/contamination values,
and save output.
"""
kwargs = _fix_uppercase(kwargs)
assert groups is None, 'Usage of LOGO in CCCV not currently implemented.'
assert not optimize, 'Parameter search over CCCV not currently implemented.'
training_records, *_ = load_training_files(
genotype_file=genotype, phenotype_file=phenotype, verb=verb
)
if params_file is not None:
loaded_params = load_params_file(params_file)
logger.info(f'Parameters loaded from file:')
logger.info('\n' + pformat(loaded_params))
kwargs = {**kwargs, **loaded_params} # TODO: should loaded params have precendence?
clf = CLF_MAPPER[type](verb=verb, *args, **kwargs)
reduce_features = True if n_features is not None else False
cccv = clf.crossvalidate_cc(records=training_records, cv=folds, n_replicates=replicates,
comple_steps=comple_steps, conta_steps=conta_steps,
n_jobs=threads, reduce_features=reduce_features,
n_features=n_features)
write_cccv_accuracy_file(out, cccv)
def test_resampling(trait_name):
td, *_ = load_training_files(
FLAT_PATH / trait_name / f'{trait_name}.genotype',
FLAT_PATH / trait_name / f'{trait_name}.phenotype')
trr = TrainingRecordResampler(random_state=2, verb=True)
trr.fit(td)
trr.get_resampled(td[0], comple=.5, conta=.5)
def generic_cv(type, genotype, phenotype, folds, replicates, threads, verb, optimize=False,
optimize_out=None, optimize_n_iter=None, groups=None, rank=None, out=None,
n_features=None, params_file=None,
*args, **kwargs):
"""
Estimate model performance by cross-validation.
Optionally, perform parameter search and save found parameters.
"""
kwargs = _fix_uppercase(kwargs)
training_records, *_ = load_training_files(
genotype_file=genotype,
phenotype_file=phenotype,
groups_file=groups,
selected_rank=rank,
verb=verb
)
if params_file is not None:
loaded_params = load_params_file(params_file)
logger.info(f'Parameters loaded from file:')
logger.info('\n' + pformat(loaded_params))
kwargs = {**kwargs, **loaded_params} # TODO: should loaded params have precendence?
clf = CLF_MAPPER[type](verb=verb, *args, **kwargs)
if optimize:
assert optimize_out is not None, 'No savepath for found parameters passed.'
logger.info(f'Optimizing parameters...')
found_params = clf.parameter_search(training_records, n_iter=optimize_n_iter)
params = {**kwargs, **found_params}
write_params_file(optimize_out, params)
logger.info(f'Optimized parameters written to file {optimize_out}.')
clf = CLF_MAPPER[type](verb=verb, *args, **params)
reduce_features = True if n_features is not None else False
use_groups = groups is not None
logger.info(f'Running CV...')
score_mean, score_sd, misclass = clf.crossvalidate(records=training_records, cv=folds,
n_replicates=replicates, groups=use_groups,
n_jobs=threads,
reduce_features=reduce_features,
n_features=n_features,
demote=not verb)
logger.info(f"CV score: {round(score_mean, 4)} +/- {round(score_sd, 4)}")
if out is not None:
write_misclassifications_file(out, training_records, misclass, use_groups=use_groups)
def generic_train(type,
genotype,
phenotype,
verb,
weights,
out,
n_features=None,
params_file=None,
*args,
**kwargs):
"""
Train and save a TrexClassifier model.
"""
kwargs = _fix_uppercase(kwargs)
training_records, *_ = load_training_files(genotype_file=genotype,
phenotype_file=phenotype,
verb=verb)
if params_file is not None:
loaded_params = load_params_file(params_file)
logger.info(f'Parameters loaded from file:')
logger.info('\n' + pformat(loaded_params))
kwargs = {**kwargs, **loaded_params}
clf = CLF_MAPPER[type](verb=verb, *args, **kwargs)
reduce_features = True if n_features is not None else False
clf.train(records=training_records,
reduce_features=reduce_features,
n_features=n_features)
if weights:
weights = clf.get_feature_weights()
weights_file_name = f"{out}.rank"
if clf.feature_type.startswith('eggNOG5'):
text_annotator = Eggnog5TextAnnotator()
feature_taxon = int(clf.feature_type.split('-')[-1])
annots = [
text_annotator.annotate(taxon_id=feature_taxon, enog_id=x)[1]
for x in weights.keys()
]
else:
annots = None
write_weights_file(weights_file=weights_file_name,
weights=weights,
annots=annots)
save_classifier(obj=clf, filename=out, verb=verb)
def test_get_shap(self, trait_name, classifier_type):
"""
Get ShapHandler and SHAP data from classifier and genotype file.
:param trait_name:
:param classifier_type:
:return:
"""
full_path_genotype = FLAT_PATH / trait_name / f"{trait_name}.genotype"
full_path_phenotype = FLAT_PATH / trait_name / f"{trait_name}.phenotype"
training_records, genotype, phenotype, group = load_training_files(
genotype_file=full_path_genotype,
phenotype_file=full_path_phenotype,
verb=True)
tr = training_records[:3]
model_path = MODELS_PATH / trait_name / f'{trait_name}.{classifier_type.lower()}.pkl'
clf = load_classifier(model_path, verb=True)
sh = ShapHandler.from_clf(clf)
fs, sv, bv = clf.get_shap(tr, n_samples=50)
return tr, sh, fs, sv, bv
def test_load_data(self, trait_name, do_write):
"""
Test training data loading. Check/catch invalid file formats.
:param trait_name:
:return:
"""
full_path_genotype = FLAT_PATH / trait_name / f"{trait_name}.genotype"
full_path_phenotype = FLAT_PATH / trait_name / f"{trait_name}.phenotype"
full_path_groups = FLAT_PATH / trait_name / f"{trait_name}.taxids"
training_records, genotype, phenotype, group = load_training_files(
genotype_file=full_path_genotype,
phenotype_file=full_path_phenotype,
groups_file=full_path_groups,
verb=True)
if do_write:
with TemporaryDirectory() as tmpdir:
gt_out = Path(tmpdir) / 'gt.genotype'
write_genotype_file(genotype, gt_out)
assert gt_out.is_file()
return training_records, genotype, phenotype, group
def generic_train(type, genotype, phenotype, verb, weights, out,
n_features=None, params_file=None, *args, **kwargs):
"""
Train and save a TrexClassifier model.
"""
kwargs = _fix_uppercase(kwargs)
training_records, *_ = load_training_files(
genotype_file=genotype, phenotype_file=phenotype, verb=verb
)
if params_file is not None:
loaded_params = load_params_file(params_file)
logger.info(f'Parameters loaded from file:')
logger.info('\n' + pformat(loaded_params))
kwargs = {**kwargs, **loaded_params}
clf = CLF_MAPPER[type](verb=verb, *args, **kwargs)
reduce_features = True if n_features is not None else False
clf.train(records=training_records, reduce_features=reduce_features, n_features=n_features)
if weights:
weights = clf.get_feature_weights()
weights_file_name = f"{out}.rank"
write_weights_file(weights_file=weights_file_name, weights=weights)
save_classifier(obj=clf, filename=out, verb=verb)
def get_training_data(self):
td, *_ = load_training_files(
FLAT_PATH/trait_name/f'{trait_name}.genotype',
FLAT_PATH/trait_name/f'{trait_name}.phenotype'
)
return td