def test(self):
path = EnvironmentSettings.tmp_test_path / "integration_sequence_classification/"
dataset = RandomDatasetGenerator.generate_sequence_dataset(50, {4: 1}, {'l1': {1: 0.5, 2: 0.5}}, path / 'data')
os.environ["cache_type"] = "test"
encoder_params = {
"normalization_type": NormalizationType.RELATIVE_FREQUENCY.name,
"reads": ReadsType.UNIQUE.name,
"sequence_encoding": SequenceEncodingType.CONTINUOUS_KMER.name,
"sequence_type": SequenceType.AMINO_ACID.name,
"k": 3
}
hp_setting = HPSetting(encoder=KmerFrequencyEncoder.build_object(dataset, **encoder_params), encoder_params=encoder_params,
ml_method=LogisticRegression(), ml_params={"model_selection_cv": False, "model_selection_n_folds": -1},
preproc_sequence=[])
lc = LabelConfiguration()
lc.add_label("l1", [1, 2])
instruction = TrainMLModelInstruction(dataset, GridSearch([hp_setting]), [hp_setting],
SplitConfig(SplitType.RANDOM, 1, 0.5, reports=ReportConfig()),
SplitConfig(SplitType.RANDOM, 1, 0.5, reports=ReportConfig()),
{Metric.BALANCED_ACCURACY}, Metric.BALANCED_ACCURACY, lc, path)
result = instruction.run(result_path=path)
shutil.rmtree(path)
def test_run(self):
path = EnvironmentSettings.root_path / "test/tmp/smmodel/"
PathBuilder.build(path)
repertoires, metadata = RepertoireBuilder.build(
[["AAA", "CCC"], ["TTTT"], ["AAA", "CCC"],
["TTTT"], ["AAA", "CCC"], ["TTTT"], ["AAA", "CCC"], ["TTTT"],
["AAA", "CCC"], ["TTTT"], ["AAA", "CCC"],
["TTTT"], ["AAA", "CCC"], ["TTTT"], ["AAA", "CCC"], ["TTTT"],
["AAA", "CCC"], ["TTTT"], ["AAA", "CCC"],
["TTTT"], ["AAA", "CCC"], ["TTTT"], ["AAA", "CCC"], ["TTTT"],
["AAA", "CCC"], ["TTTT"], ["AAA", "CCC"], ["TTTT"],
["AAA", "CCC"], ["TTTT"], ["AAA", "CCC"], ["TTTT"]], path, {
"default": [
1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1,
2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2
]
})
dataset = RepertoireDataset(repertoires=repertoires,
labels={"default": [1, 2]},
metadata_file=metadata)
label_config = LabelConfiguration()
label_config.add_label("default", [1, 2])
hp_settings = [
HPSetting(
Word2VecEncoder.build_object(
dataset, **{
"vector_size": 8,
"model_type": ModelType.SEQUENCE.name,
"k": 3
}), {
"vector_size": 8,
"model_type": ModelType.SEQUENCE.name,
"k": 3
}, LogisticRegression(), {
"model_selection_cv": False,
"model_selection_n_folds": -1
}, [])
]
split_config_assessment = SplitConfig(SplitType.RANDOM, 1, 0.5,
ReportConfig())
split_config_selection = SplitConfig(SplitType.RANDOM, 1, 0.5,
ReportConfig())
instruction = TrainMLModelInstruction(
dataset, GridSearch(hp_settings), hp_settings,
split_config_assessment, split_config_selection,
{Metric.BALANCED_ACCURACY}, Metric.BALANCED_ACCURACY, label_config,
path)
semantic_model = SemanticModel([instruction], path)
semantic_model.run()
shutil.rmtree(path)
def _update_split_configs(self, assessment: SplitConfig, selection: SplitConfig, dataset: Dataset) -> Tuple[SplitConfig, SplitConfig]:
if assessment.split_strategy == SplitType.LOOCV:
assessment.split_count = dataset.get_example_count()
train_val_example_count = assessment.split_count - 1
elif assessment.split_strategy == SplitType.K_FOLD:
train_val_example_count = int(dataset.get_example_count() * (assessment.split_count - 1) / assessment.split_count)
else:
train_val_example_count = int(dataset.get_example_count() * assessment.training_percentage)
if selection.split_strategy == SplitType.LOOCV:
selection.split_count = train_val_example_count
return assessment, selection
def test__split_repertoire_dataset(self):
path = PathBuilder.build(EnvironmentSettings.tmp_test_path / "manual_splitter/")
dataset = RandomDatasetGenerator.generate_repertoire_dataset(10, {4: 1}, {3: 1}, {}, path)
train_metadata = pd.DataFrame({"subject_id": ["rep_1", "rep_2", "rep_4", "rep_5", "rep_9", "rep_7"]})
train_metadata.to_csv(path / "train.csv")
test_metadata = pd.DataFrame({"subject_id": ["rep_0", "rep_3", "rep_6", "rep_8"]})
test_metadata.to_csv(path / "test.csv")
train_datasets, test_datasets = ManualSplitter._split_repertoire_dataset(
DataSplitterParams(dataset, SplitType.MANUAL, split_count=1, paths=[path / 'result/'],
split_config=SplitConfig(manual_config=ManualSplitConfig(path / "train.csv",
path / "test.csv"),
split_count=1, split_strategy=SplitType.MANUAL)))
self.assertEqual(1, len(train_datasets))
self.assertEqual(1, len(test_datasets))
self.assertEqual(6, train_datasets[0].get_example_count())
self.assertEqual(4, test_datasets[0].get_example_count())
self.assertTrue(all(subject_id in ["rep_1", "rep_2", "rep_4", "rep_5", "rep_9", "rep_7"]
for subject_id in train_datasets[0].get_metadata(["subject_id"])["subject_id"]))
self.assertTrue(all(subject_id in ["rep_0", "rep_3", "rep_6", "rep_8"]
for subject_id in test_datasets[0].get_metadata(["subject_id"])["subject_id"]))
self.assertTrue(os.path.isfile(train_datasets[0].metadata_file))
self.assertTrue(os.path.isfile(test_datasets[0].metadata_file))
shutil.rmtree(path)
def _parse_split_config(self, instruction_key, instruction: dict, split_key: str, symbol_table: SymbolTable, settings_count: int) -> SplitConfig:
try:
default_params = DefaultParamsLoader.load("instructions/", SplitConfig.__name__)
report_config_input = self._prepare_report_config(instruction_key, instruction, split_key, symbol_table)
instruction[split_key] = {**default_params, **instruction[split_key]}
split_strategy = SplitType[instruction[split_key]["split_strategy"].upper()]
training_percentage = float(instruction[split_key]["training_percentage"]) if split_strategy == SplitType.RANDOM else -1
if split_strategy == SplitType.RANDOM and training_percentage == 1 and settings_count > 1:
raise ValueError(f"{TrainMLModelParser.__name__}: all data under {instruction_key}/{split_key} was specified to be used for "
f"training, but {settings_count} settings were specified for evaluation. Please define a test/validation set by "
f"reducing the training percentage (e.g., to 0.7) or use only one hyperparameter setting to run the analysis.")
return SplitConfig(split_strategy=split_strategy,
split_count=int(instruction[split_key]["split_count"]),
training_percentage=training_percentage,
reports=ReportConfig(**report_config_input),
manual_config=ManualSplitConfig(**instruction[split_key]["manual_config"]) if "manual_config" in instruction[split_key] else None,
leave_one_out_config=LeaveOneOutConfig(**instruction[split_key]["leave_one_out_config"])
if "leave_one_out_config" in instruction[split_key] else None)
except KeyError as key_error:
raise KeyError(f"{TrainMLModelParser.__name__}: parameter {key_error.args[0]} was not defined under {split_key}.")
def _create_state_object(self, path):
repertoires, metadata = RepertoireBuilder.build(sequences=[["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"]],
path=path,
labels={
"l1": [1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2,
1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2],
"l2": [0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 1,
0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1]})
dataset = RepertoireDataset(repertoires=repertoires, metadata_file=metadata,
labels={"l1": [1, 2], "l2": [0, 1]})
enc_params = {"k": 3, "model_type": ModelType.SEQUENCE.name, "vector_size": 4}
hp_settings = [HPSetting(Word2VecEncoder.build_object(dataset, **enc_params), enc_params,
LogisticRegression(),
{"model_selection_cv": False, "model_selection_n_folds": -1},
[])]
label_config = LabelConfiguration([Label("l1", [1, 2]), Label("l2", [0, 1])])
process = TrainMLModelInstruction(dataset, GridSearch(hp_settings), hp_settings,
SplitConfig(SplitType.RANDOM, 1, 0.7),
SplitConfig(SplitType.RANDOM, 1, 0.7),
{Metric.BALANCED_ACCURACY}, Metric.BALANCED_ACCURACY, label_config, path)
state = process.run(result_path=path)
return state
def test(self):
path = EnvironmentSettings.tmp_test_path / "integration_receptor_classification/"
dataset = self.create_dataset(path)
os.environ["cache_type"] = "test"
encoder_params = {
"normalization_type": NormalizationType.RELATIVE_FREQUENCY.name,
"reads": ReadsType.UNIQUE.name,
"sequence_encoding": SequenceEncodingType.CONTINUOUS_KMER.name,
"sequence_type": SequenceType.AMINO_ACID.name,
"k": 3
}
hp_setting = HPSetting(encoder=KmerFrequencyEncoder.build_object(
dataset, **encoder_params),
encoder_params=encoder_params,
ml_method=LogisticRegression(),
ml_params={
"model_selection_cv": False,
"model_selection_n_folds": -1
},
preproc_sequence=[])
lc = LabelConfiguration()
lc.add_label("l1", [1, 2])
instruction = TrainMLModelInstruction(
dataset, GridSearch([hp_setting]), [hp_setting],
SplitConfig(SplitType.RANDOM, 1, 0.5, reports=ReportConfig()),
SplitConfig(SplitType.RANDOM, 1, 0.5, reports=ReportConfig()),
{Metric.BALANCED_ACCURACY}, Metric.BALANCED_ACCURACY, lc, path)
state = instruction.run(result_path=path)
print(vars(state))
self.assertEqual(
1.0, state.assessment_states[0].label_states["l1"].
optimal_assessment_item.performance[
state.optimization_metric.name.lower()])
shutil.rmtree(path)
def test_split_dataset(self):
path = PathBuilder.build(EnvironmentSettings.tmp_test_path /
"leave_one_out_splitter/")
receptors = []
for i in range(10):
receptors.append(
TCABReceptor(ReceptorSequence(), ReceptorSequence(),
{"subject": i % 3}))
filename = path / "batch1.pickle"
with open(filename, "wb") as file:
pickle.dump(receptors, file)
dataset = ReceptorDataset(filenames=[filename])
params = DataSplitterParams(
dataset,
SplitType.LEAVE_ONE_OUT_STRATIFICATION,
3,
paths=[path / f"result_{i}/" for i in range(1, 4)],
split_config=SplitConfig(SplitType.LEAVE_ONE_OUT_STRATIFICATION,
split_count=3,
leave_one_out_config=LeaveOneOutConfig(
"subject", 1)))
train_datasets, test_datasets = LeaveOneOutSplitter.split_dataset(
params)
self.assertEqual(3, len(train_datasets))
self.assertEqual(3, len(test_datasets))
for i in range(3):
self.assertTrue(
all(receptor.metadata["subject"] == i
for receptor in test_datasets[i].get_data()))
self.assertTrue(
all(receptor.metadata["subject"] != i
for receptor in train_datasets[i].get_data()))
shutil.rmtree(path)
def test_generate(self):
path = EnvironmentSettings.tmp_test_path / "cv_feature_performance/"
state = TrainMLModelState(
assessment=SplitConfig(split_count=5,
split_strategy=SplitType.K_FOLD),
selection=SplitConfig(split_count=10,
split_strategy=SplitType.K_FOLD),
optimization_metric=Metric.ACCURACY,
label_configuration=LabelConfiguration(
labels=[Label(name="CMV", values=[True, False])]),
hp_settings=[
HPSetting(encoder_params={"p_value_threshold": 0.001},
encoder_name="e1",
encoder=SequenceAbundanceEncoder([], 0, 0, 0),
preproc_sequence=[],
ml_method_name="ml1",
ml_method=ProbabilisticBinaryClassifier(10, 0.1),
ml_params={}),
HPSetting(encoder_params={"p_value_threshold": 0.01},
encoder_name="e2",
encoder=SequenceAbundanceEncoder([], 0, 0, 0),
preproc_sequence=[],
ml_method_name="ml1",
ml_method=ProbabilisticBinaryClassifier(10, 0.1),
ml_params={}),
HPSetting(encoder_params={"p_value_threshold": 0.01},
encoder=SequenceAbundanceEncoder([], 0, 0, 0),
preproc_sequence=[],
ml_method=ProbabilisticBinaryClassifier(10, 0.01),
ml_params={})
],
dataset=None,
hp_strategy=None,
metrics=None)
report = CVFeaturePerformance("p_value_threshold",
state,
path,
is_feature_axis_categorical=True,
name="report1")
with self.assertWarns(RuntimeWarning):
report.generate_report()
state.hp_settings = state.hp_settings[:2]
state.assessment_states = [
HPAssessmentState(i, None, None, None, state.label_configuration)
for i in range(state.assessment.split_count)
]
for assessment_state in state.assessment_states:
assessment_state.label_states["CMV"] = HPLabelState("CMV", [])
assessment_state.label_states["CMV"].assessment_items = {
setting.get_key():
HPItem(performance={'accuracy': random.uniform(0.5, 1)},
hp_setting=setting)
for setting in state.hp_settings
}
assessment_state.label_states[
"CMV"].selection_state = HPSelectionState(
[], [], "", GridSearch(state.hp_settings))
assessment_state.label_states["CMV"].selection_state.hp_items = {
str(setting): [
HPItem(performance={'accuracy': random.uniform(0.5, 1)},
hp_setting=setting)
for _ in range(state.selection.split_count)
]
for setting in state.hp_settings
}
report.state = state
report_result = report.generate_report()
self.assertTrue(isinstance(report_result, ReportResult))
self.assertEqual(2, len(report_result.output_tables))
self.assertEqual(1, len(report_result.output_figures))
self.assertTrue(os.path.isfile(report_result.output_figures[0].path))
self.assertTrue(os.path.isfile(report_result.output_tables[0].path))
self.assertTrue(os.path.isfile(report_result.output_tables[1].path))
shutil.rmtree(path)
def test_run(self):
path = EnvironmentSettings.tmp_test_path / "hpoptimproc/"
PathBuilder.build(path)
repertoires, metadata = RepertoireBuilder.build(
sequences=[["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"],
["AAA", "CCC", "DDD"], ["AAA", "CCC", "DDD"]],
path=path,
labels={
"l1": [
1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2,
1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2
],
"l2": [
0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 1, 0, 1,
0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1
]
})
dataset = RepertoireDataset(repertoires=repertoires,
metadata_file=metadata,
labels={
"l1": [1, 2],
"l2": [0, 1]
})
enc1 = {
"k": 3,
"model_type": ModelType.SEQUENCE.name,
"vector_size": 4
}
enc2 = {
"k": 3,
"model_type": ModelType.SEQUENCE.name,
"vector_size": 6
}
hp_settings = [
HPSetting(Word2VecEncoder.build_object(dataset, **enc1), enc1,
LogisticRegression(), {
"model_selection_cv": False,
"model_selection_n_folds": -1
}, []),
HPSetting(
Word2VecEncoder.build_object(dataset, **enc2), enc2, SVM(), {
"model_selection_cv": False,
"model_selection_n_folds": -1
},
[ClonesPerRepertoireFilter(lower_limit=-1, upper_limit=1000)])
]
report = SequenceLengthDistribution()
label_config = LabelConfiguration(
[Label("l1", [1, 2]), Label("l2", [0, 1])])
process = TrainMLModelInstruction(
dataset, GridSearch(hp_settings), hp_settings,
SplitConfig(SplitType.RANDOM,
1,
0.5,
reports=ReportConfig(data_splits={"seqlen": report})),
SplitConfig(SplitType.RANDOM,
1,
0.5,
reports=ReportConfig(data_splits={"seqlen": report})),
{Metric.BALANCED_ACCURACY}, Metric.BALANCED_ACCURACY, label_config,
path)
state = process.run(result_path=path)
self.assertTrue(isinstance(state, TrainMLModelState))
self.assertEqual(1, len(state.assessment_states))
self.assertTrue("l1" in state.assessment_states[0].label_states)
self.assertTrue("l2" in state.assessment_states[0].label_states)
shutil.rmtree(path)
