From a9bac0c6c618f7f1fca7c20d357582ec31717719 Mon Sep 17 00:00:00 2001
From: Tim O'Donnell <timodonnell@gmail.com>
Date: Wed, 1 Mar 2017 21:34:04 -0500
Subject: [PATCH] fixes
---
.../class1_ensemble_multi_allele_predictor.py | 82 +++++++++----------
test/test_ensemble.py | 6 --
2 files changed, 39 insertions(+), 49 deletions(-)
diff --git a/mhcflurry/class1_allele_specific_ensemble/class1_ensemble_multi_allele_predictor.py b/mhcflurry/class1_allele_specific_ensemble/class1_ensemble_multi_allele_predictor.py
index 8c23c6d4..19021c35 100644
--- a/mhcflurry/class1_allele_specific_ensemble/class1_ensemble_multi_allele_predictor.py
+++ b/mhcflurry/class1_allele_specific_ensemble/class1_ensemble_multi_allele_predictor.py
@@ -45,7 +45,7 @@ IMPUTE_HYPERPARAMETER_DEFAULTS = HyperparameterDefaults(
impute_method='mice',
impute_min_observations_per_peptide=1,
impute_min_observations_per_allele=1,
- imputer_args=[{"n_burn_in": 5, "n_imputations": 25}])
+ imputer_args={"n_burn_in": 5, "n_imputations": 25})
HYPERPARAMETER_DEFAULTS = (
HyperparameterDefaults(
@@ -64,14 +64,13 @@ def fit_and_test(
fold_num,
train_measurement_collection_broadcast,
test_measurement_collection_broadcast,
- alleles,
- hyperparameters_list):
+ allele_and_hyperparameter_pairs):
assert len(train_measurement_collection_broadcast.value.df) > 0
assert len(test_measurement_collection_broadcast.value.df) > 0
results = []
- for all_hyperparameters in hyperparameters_list:
+ for (allele, all_hyperparameters) in allele_and_hyperparameter_pairs:
measurement_collection_hyperparameters = (
MEASUREMENT_COLLECTION_HYPERPARAMETER_DEFAULTS.subselect(
all_hyperparameters))
@@ -79,34 +78,33 @@ def fit_and_test(
Class1BindingPredictor.hyperparameter_defaults.subselect(
all_hyperparameters))
- for allele in alleles:
- train_dataset = (
- train_measurement_collection_broadcast
- .value
- .select_allele(allele)
- .to_dataset(**measurement_collection_hyperparameters))
- test_dataset = (
- test_measurement_collection_broadcast
- .value
- .select_allele(allele)
- .to_dataset(**measurement_collection_hyperparameters))
-
- assert len(train_dataset) > 0
- assert len(test_dataset) > 0
-
- model = Class1BindingPredictor(**model_hyperparameters)
-
- model.fit_dataset(train_dataset)
- predictions = model.predict(test_dataset.peptides)
- scores = scoring.make_scores(
- test_dataset.affinities, predictions)
- results.append({
- 'fold_num': fold_num,
- 'allele': allele,
- 'hyperparameters': all_hyperparameters,
- 'model': model,
- 'scores': scores
- })
+ train_dataset = (
+ train_measurement_collection_broadcast
+ .value
+ .select_allele(allele)
+ .to_dataset(**measurement_collection_hyperparameters))
+ test_dataset = (
+ test_measurement_collection_broadcast
+ .value
+ .select_allele(allele)
+ .to_dataset(**measurement_collection_hyperparameters))
+
+ assert len(train_dataset) > 0
+ assert len(test_dataset) > 0
+
+ model = Class1BindingPredictor(**model_hyperparameters)
+
+ model.fit_dataset(train_dataset)
+ predictions = model.predict(test_dataset.peptides)
+ scores = scoring.make_scores(
+ test_dataset.affinities, predictions)
+ results.append({
+ 'fold_num': fold_num,
+ 'allele': allele,
+ 'hyperparameters': all_hyperparameters,
+ 'model': model,
+ 'scores': scores
+ })
return results
@@ -321,19 +319,16 @@ class Class1EnsembleMultiAllelePredictor(object):
train_broadcast = parallel_backend.broadcast(train_split)
test_broadcast = parallel_backend.broadcast(test_split)
- task_alleles = []
- task_models = []
+ task_allele_model_pairs = []
def make_task():
- if task_alleles and task_models:
+ if task_allele_model_pairs:
tasks.append((
fold_num,
train_broadcast,
test_broadcast,
- list(task_alleles),
- list(task_models)))
- task_alleles.clear()
- task_models.clear()
+ list(task_allele_model_pairs)))
+ task_allele_model_pairs.clear()
assert all(
allele in set(train_split.df.allele.unique())
@@ -348,12 +343,11 @@ class Class1EnsembleMultiAllelePredictor(object):
for allele in alleles:
for model in self.hyperparameters_to_search:
- task_models.append(model)
- if len(task_alleles) * len(task_models) > work_per_task:
+ task_allele_model_pairs.append((allele, model))
+ if len(task_allele_model_pairs) > work_per_task:
make_task()
make_task()
- assert not task_alleles
- assert not task_models
+ assert not task_allele_model_pairs
logging.info(
"Training and scoring models: %d tasks (target was %d), "
@@ -365,6 +359,7 @@ class Class1EnsembleMultiAllelePredictor(object):
len(self.hyperparameters_to_search),
total_work))
+ assert len(tasks) > 0
results = parallel_backend.map(call_fit_and_test, tasks)
# fold number -> allele -> best model
@@ -404,6 +399,7 @@ class Class1EnsembleMultiAllelePredictor(object):
manifest_entry["%s_%s" % (key, sub_key)] = value
manifest_rows.append(manifest_entry)
+ assert len(manifest_rows) > 0
manifest_df = pandas.DataFrame(manifest_rows)
manifest_df.index = manifest_df.model_name
del manifest_df["model_name"]
diff --git a/test/test_ensemble.py b/test/test_ensemble.py
index 099aeaae..275d5bf9 100644
--- a/test/test_ensemble.py
+++ b/test/test_ensemble.py
@@ -21,12 +21,6 @@ from mhcflurry \
Class1EnsembleMultiAllelePredictor,
HYPERPARAMETER_DEFAULTS)
-try:
- import kubeface
- KUBEFACE_INSTALLED = True
-except ImportError:
- KUBEFACE_INSTALLED = False
-
def test_basic():
model_hyperparameters = HYPERPARAMETER_DEFAULTS.models_grid(
--
GitLab