diff --git a/mhcflurry/__init__.py b/mhcflurry/__init__.py
index 4420dcbbfd919085e5bb9e82d0573f4ad11969a4..dbb0b2e2aeac63fa5588026c502efeb2643853f3 100644
--- a/mhcflurry/__init__.py
+++ b/mhcflurry/__init__.py
@@ -12,15 +12,15 @@
# See the License for the specific language governing permissions and
# limitations under the License.
-from .class1_affinity_prediction.class1_binding_predictor import (
- Class1BindingPredictor)
-from .class1_affinity_prediction.multi_allele_predictor_ensemble import (
- MultiAllelePredictorEnsemble)
+from .class1_affinity_prediction.class1_neural_network import (
+ Class1NeuralNetwork)
+from .class1_affinity_prediction.class1_affinity_predictor import (
+ Class1AffinityPredictor)
__version__ = "0.2.0"
__all__ = [
- "Class1BindingPredictor",
- "MultiAllelePredictorEnsemble",
+ "Class1NeuralNetwork",
+ "Class1AffinityPredictor",
"__version__",
]
diff --git a/mhcflurry/antigen_presentation/presentation_component_models/mhcflurry_trained_on_hits.py b/mhcflurry/antigen_presentation/presentation_component_models/mhcflurry_trained_on_hits.py
index 58d0101e32535a48083b26d23bd3243d547237ea..9506e4e45ae2adb42ecf553df3622baeb900bdee 100644
--- a/mhcflurry/antigen_presentation/presentation_component_models/mhcflurry_trained_on_hits.py
+++ b/mhcflurry/antigen_presentation/presentation_component_models/mhcflurry_trained_on_hits.py
@@ -4,7 +4,7 @@ import pandas
from numpy import log, exp, nanmean
from ...affinity_measurement_dataset import AffinityMeasurementDataset
-from ...class1_affinity_prediction import Class1BindingPredictor
+from ...class1_affinity_prediction import Class1NeuralNetwork
from ...common import normalize_allele_name
from .mhc_binding_component_model_base import MHCBindingComponentModelBase
@@ -99,7 +99,7 @@ class MHCflurryTrainedOnHits(MHCBindingComponentModelBase):
dataset = AffinityMeasurementDataset(
df.sample(frac=1)) # shuffle dataframe
print("Train data: ", dataset)
- model = Class1BindingPredictor(
+ model = Class1NeuralNetwork(
**self.mhcflurry_hyperparameters)
model.fit_dataset(dataset, verbose=True)
self.allele_to_model[allele] = model
diff --git a/mhcflurry/class1_affinity_prediction/__init__.py b/mhcflurry/class1_affinity_prediction/__init__.py
index 7deec3d17cba797bc79e69834d742929cf03d476..96707e88331792c6a4cc0d0214ef656b7522bfff 100644
--- a/mhcflurry/class1_affinity_prediction/__init__.py
+++ b/mhcflurry/class1_affinity_prediction/__init__.py
@@ -1,7 +1,9 @@
from __future__ import absolute_import
-from .class1_binding_predictor import Class1BindingPredictor
+from .class1_neural_network import Class1NeuralNetwork
+from .class1_affinity_predictor import Class1AffinityPredictor
__all__ = [
- 'Class1BindingPredictor',
+ 'Class1NeuralNetwork',
+ 'Class1AffinityPredictor',
]
diff --git a/mhcflurry/class1_affinity_prediction/multi_allele_predictor_ensemble.py b/mhcflurry/class1_affinity_prediction/class1_affinity_predictor.py
similarity index 98%
rename from mhcflurry/class1_affinity_prediction/multi_allele_predictor_ensemble.py
rename to mhcflurry/class1_affinity_prediction/class1_affinity_predictor.py
index 4e8b965ca4ade19be20f44023591931fefbcdb54..7c6fb5ef88c35cbf4a7832882f8217ca359b4b9f 100644
--- a/mhcflurry/class1_affinity_prediction/multi_allele_predictor_ensemble.py
+++ b/mhcflurry/class1_affinity_prediction/class1_affinity_predictor.py
@@ -11,10 +11,10 @@ import mhcnames
from ..encodable_sequences import EncodableSequences
-from .class1_binding_predictor import Class1BindingPredictor
+from .class1_neural_network import Class1NeuralNetwork
-class MultiAllelePredictorEnsemble(object):
+class Class1AffinityPredictor(object):
def __init__(
self,
allele_to_allele_specific_models={},
@@ -117,7 +117,7 @@ class MultiAllelePredictorEnsemble(object):
for (allele, v)
in sorted(allele_to_allele_specific_models.items()))))
- result = MultiAllelePredictorEnsemble(
+ result = Class1AffinityPredictor(
allele_to_allele_specific_models=allele_to_allele_specific_models,
class1_pan_allele_models=class1_pan_allele_models,
allele_to_pseudosequence=pseudosequences,
@@ -226,7 +226,7 @@ class MultiAllelePredictorEnsemble(object):
output_assignments = ["output"] * len(encodable_peptides.sequences)
for i in range(n_models):
print("Training model %d / %d" % (i + 1, n_models))
- model = Class1BindingPredictor(**architecture_hyperparameters)
+ model = Class1NeuralNetwork(**architecture_hyperparameters)
model.fit(
encodable_peptides,
affinities,
diff --git a/mhcflurry/class1_affinity_prediction/class1_binding_predictor.py b/mhcflurry/class1_affinity_prediction/class1_neural_network.py
similarity index 99%
rename from mhcflurry/class1_affinity_prediction/class1_binding_predictor.py
rename to mhcflurry/class1_affinity_prediction/class1_neural_network.py
index 04b87d065f8d25abd29b24fafad6689f7a7ecd81..df8e767c46060fbef92a77cecb2b4d3847972529 100644
--- a/mhcflurry/class1_affinity_prediction/class1_binding_predictor.py
+++ b/mhcflurry/class1_affinity_prediction/class1_neural_network.py
@@ -21,7 +21,7 @@ from ..regression_target import to_ic50, from_ic50
from ..common import random_peptides, amino_acid_distribution
-class Class1BindingPredictor(object):
+class Class1NeuralNetwork(object):
network_hyperparameter_defaults = HyperparameterDefaults(
kmer_size=15,
use_embedding=True,
diff --git a/mhcflurry/class1_affinity_prediction/cv_and_train_command.py b/mhcflurry/class1_affinity_prediction/cv_and_train_command.py
deleted file mode 100644
index 3d0159f4b1d1f6bb3c37783a92dae7a89fe40bd2..0000000000000000000000000000000000000000
--- a/mhcflurry/class1_affinity_prediction/cv_and_train_command.py
+++ /dev/null
@@ -1,400 +0,0 @@
-# Copyright (c) 2016. Mount Sinai School of Medicine
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-'''
-Class1 allele-specific cross validation and training script.
-
-What it does:
- * Run cross validation on a dataset over the specified model architectures
- * Select the best architecture for each allele
- * Re-train the best architecture on the full data for that allele
- * Test "production" predictors on a held-out test set if available
-
-Features:
- * Supports imputation as a hyperparameter that can be searched over
- * Parallelized with concurrent.futures
-
-Note:
-
-The parallelization is primary intended to be used with an
-alternative concurrent.futures Executor such as dask-distributed that supports
-multi-node parallelization. Theano in particular seems to have deadlocks
-when running with single-node parallelization.
-'''
-from __future__ import (
- print_function,
- division,
- absolute_import,
-)
-import sys
-import argparse
-import json
-import logging
-import time
-import os
-import socket
-import hashlib
-import pickle
-
-import numpy
-
-from .. import parallelism
-from ..affinity_measurement_dataset import AffinityMeasurementDataset
-from ..imputation_helpers import imputer_from_name
-from .cross_validation import cross_validation_folds
-from .train import (
- impute_and_select_allele,
- train_across_models_and_folds,
- AlleleSpecificTrainTestFold)
-
-parser = argparse.ArgumentParser(
- description=__doc__,
- formatter_class=argparse.RawDescriptionHelpFormatter)
-
-parser.add_argument(
- "--train-data",
- metavar="X.csv",
- required=True,
- help="Training data")
-
-parser.add_argument(
- "--test-data",
- metavar="X.csv",
- help="Optional test data")
-
-parser.add_argument(
- "--model-architectures",
- metavar="X.json",
- type=argparse.FileType('r'),
- required=True,
- help="JSON file giving model architectures to assess in cross validation."
- " Can be - to read from stdin")
-
-parser.add_argument(
- "--imputer-description",
- metavar="X.json",
- type=argparse.FileType('r'),
- help="JSON. Can be - to read from stdin")
-
-parser.add_argument(
- "--alleles",
- metavar="ALLELE",
- nargs="+",
- default=None,
- help="Use only the specified alleles")
-
-parser.add_argument(
- "--out-cv-results",
- metavar="X.csv",
- help="Write cross validation results to the given file")
-
-parser.add_argument(
- "--out-production-results",
- metavar="X.csv",
- help="Write production model information to the given file")
-
-parser.add_argument(
- "--out-models-dir",
- metavar="DIR",
- help="Write production models to files in this dir")
-
-parser.add_argument(
- "--max-models",
- type=int,
- metavar="N",
- help="Use only the first N models")
-
-parser.add_argument(
- "--cv-num-folds",
- type=int,
- default=3,
- metavar="N",
- help="Number of cross validation folds. Default: %(default)s")
-
-parser.add_argument(
- "--cv-folds-per-task",
- type=int,
- default=10,
- metavar="N",
- help="When parallelizing cross validation, each task trains one model "
- "architecture on N folds. Set to 1 for maximum potential parallelism. "
- "This is less efficient if you have limited workers, however, since "
- "the model must get compiled for each task. Default: %(default)s.")
-
-parser.add_argument(
- "--dask-scheduler",
- metavar="HOST:PORT",
- help="Host and port of dask distributed scheduler")
-
-parser.add_argument(
- "--num-local-processes",
- metavar="N",
- type=int,
- help="Processes (exclusive with --dask-scheduler and --num-local-threads)")
-
-parser.add_argument(
- "--num-local-threads",
- metavar="N",
- type=int,
- default=1,
- help="Threads (exclusive with --dask-scheduler and --num-local-processes)")
-
-parser.add_argument(
- "--min-samples-per-allele",
- default=100,
- metavar="N",
- help="Don't train predictors for alleles with fewer than N samples. "
- "Set to 0 to disable filtering. Default: %(default)s",
- type=int)
-
-parser.add_argument(
- "--quiet",
- action="store_true",
- default=False,
- help="Output less info")
-
-parser.add_argument(
- "--verbose",
- action="store_true",
- default=False,
- help="Output more info")
-
-try:
- import kubeface
- kubeface.Client.add_args(parser)
-except ImportError:
- logging.error("Kubeface support disabled, not installed.")
-
-
-def run(argv=sys.argv[1:]):
- args = parser.parse_args(argv)
- if args.verbose:
- logging.root.setLevel(level="DEBUG")
- elif not args.quiet:
- logging.root.setLevel(level="INFO")
-
- logging.info("Running with arguments: %s" % args)
-
- # Set parallel backend
- if args.dask_scheduler:
- backend = parallelism.DaskDistributedParallelBackend(
- args.dask_scheduler)
- elif hasattr(args, 'storage_prefix') and args.storage_prefix:
- backend = parallelism.KubefaceParallelBackend(args)
- else:
- if args.num_local_processes:
- backend = parallelism.ConcurrentFuturesParallelBackend(
- args.num_local_processes,
- processes=True)
- else:
- backend = parallelism.ConcurrentFuturesParallelBackend(
- args.num_local_threads,
- processes=False)
-
- parallelism.set_default_backend(backend)
- logging.info("Using parallel backend: %s" % backend)
- go(args)
-
-
-def go(args):
- backend = parallelism.get_default_backend()
-
- model_architectures = json.loads(args.model_architectures.read())
- logging.info("Read %d model architectures" % len(model_architectures))
- if args.max_models:
- model_architectures = model_architectures[:args.max_models]
- logging.info(
- "Subselected to %d model architectures" % len(model_architectures))
-
- train_data = AffinityMeasurementDataset.from_csv(args.train_data)
- logging.info("Loaded training dataset: %s" % train_data)
-
- test_data = None
- if args.test_data:
- test_data = AffinityMeasurementDataset.from_csv(args.test_data)
- logging.info("Loaded testing dataset: %s" % test_data)
-
- if args.min_samples_per_allele:
- train_data = train_data.filter_alleles_by_count(
- args.min_samples_per_allele)
- logging.info(
- "Filtered training dataset to alleles with >= %d observations: %s"
- % (args.min_samples_per_allele, train_data))
-
- if any(x['impute'] for x in model_architectures):
- if not args.imputer_description:
- parser.error(
- "--imputer-description is required when any models "
- "use imputation")
- imputer_description = json.load(args.imputer_description)
- logging.info("Loaded imputer description: %s" % imputer_description)
- imputer_kwargs_defaults = {
- 'min_observations_per_peptide': 2,
- 'min_observations_per_allele': 10,
- }
- impute_kwargs = dict(
- (key, imputer_description.pop(key, default))
- for (key, default) in imputer_kwargs_defaults.items())
-
- imputer = imputer_from_name(**imputer_description)
- else:
- imputer = None
- impute_kwargs = {}
-
- logging.info(
- "Generating cross validation folds. Imputation: %s" %
- ("yes" if imputer else "no"))
- cv_folds = cross_validation_folds(
- train_data,
- n_folds=args.cv_num_folds,
- imputer=imputer,
- impute_kwargs=impute_kwargs,
- drop_similar_peptides=True,
- alleles=args.alleles)
-
- logging.info(
- "Training %d model architectures across %d folds = %d models"
- % (
- len(model_architectures),
- len(cv_folds),
- len(model_architectures) * len(cv_folds)))
- start = time.time()
- cv_results = train_across_models_and_folds(
- cv_folds,
- model_architectures,
- folds_per_task=args.cv_folds_per_task)
- logging.info(
- "Completed cross validation in %0.2f seconds" % (time.time() - start))
-
- cv_results["summary_score"] = (
- cv_results.test_auc.fillna(0) +
- cv_results.test_tau.fillna(0) +
- cv_results.test_f1.fillna(0))
-
- allele_and_model_to_ranks = {}
- for allele in cv_results.allele.unique():
- model_ranks = (
- cv_results.ix[cv_results.allele == allele]
- .groupby("model_num")
- .summary_score
- .mean()
- .rank(method='first', ascending=False, na_option="top")
- .astype(int))
- allele_and_model_to_ranks[allele] = model_ranks.to_dict()
-
- cv_results["summary_rank"] = [
- allele_and_model_to_ranks[row.allele][row.model_num]
- for (_, row) in cv_results.iterrows()
- ]
-
- if args.out_cv_results:
- cv_results.to_csv(args.out_cv_results, index=False)
- print("Wrote: %s" % args.out_cv_results)
-
- numpy.testing.assert_equal(
- set(cv_results.summary_rank),
- set(1 + numpy.arange(len(model_architectures))))
-
- best_architectures_by_allele = (
- cv_results.ix[cv_results.summary_rank == 1]
- .set_index("allele")
- .model_num
- .to_dict())
-
- logging.info("")
- train_folds = []
- train_models = []
- imputation_args_list = []
- best_architectures = []
- for (allele_num, allele) in enumerate(cv_results.allele.unique()):
- best_index = best_architectures_by_allele[allele]
- architecture = model_architectures[best_index]
- best_architectures.append(architecture)
- train_models.append(architecture)
- logging.info(
- "Allele: %s best architecture is index %d: %s" %
- (allele, best_index, architecture))
-
- if architecture['impute']:
- imputation_args = dict(impute_kwargs)
- imputation_args.update(dict(
- dataset=train_data,
- imputer=imputer,
- allele=allele))
- imputation_args_list.append(imputation_args)
-
- test_data_this_allele = None
- if test_data is not None:
- test_data_this_allele = test_data.get_allele(allele)
- fold = AlleleSpecificTrainTestFold(
- allele=allele,
- train=train_data.get_allele(allele),
- imputed_train=None,
- test=test_data_this_allele)
- train_folds.append(fold)
-
- if imputation_args_list:
- imputation_results = list(backend.map(
- lambda kwargs: impute_and_select_allele(**kwargs),
- imputation_args_list))
-
- new_train_folds = []
- for (best_architecture, train_fold) in zip(
- best_architectures, train_folds):
- imputed_train = None
- if best_architecture['impute']:
- imputed_train = imputation_results.pop(0)
- new_train_folds.append(
- train_fold._replace(imputed_train=imputed_train))
- assert not imputation_results
-
- train_folds = new_train_folds
-
- logging.info("Training %d production models" % len(train_folds))
- start = time.time()
- train_results = train_across_models_and_folds(
- train_folds,
- train_models,
- cartesian_product_of_folds_and_models=False,
- return_predictors=args.out_models_dir is not None)
- logging.info(
- "Completed production training in %0.2f seconds"
- % (time.time() - start))
-
- if args.out_models_dir:
- predictor_names = []
- run_name = (hashlib.sha1(
- ("%s-%f" % (socket.gethostname(), time.time())).encode())
- .hexdigest()[:8])
- for (_, row) in train_results.iterrows():
- predictor_name = "-".join(str(x) for x in [
- row.allele,
- "impute" if row.model_impute else "noimpute",
- "then".join(str(s) for s in row.model_layer_sizes),
- "dropout%g" % row.model_dropout_probability,
- "fracneg%g" % row.model_fraction_negative,
- run_name,
- ]).replace(".", "_")
- predictor_names.append(predictor_name)
- out_path = os.path.join(
- args.out_models_dir, predictor_name + ".pickle")
- with open(out_path, "wb") as fd:
- # Use this protocol so we have Python 2 compatability.
- pickle.dump(row.predictor, fd, protocol=2)
- print("Wrote: %s" % out_path)
- del train_results["predictor"]
- train_results["predictor_name"] = predictor_names
-
- if args.out_production_results:
- train_results.to_csv(args.out_production_results, index=False)
- print("Wrote: %s" % args.out_production_results)
diff --git a/mhcflurry/class1_affinity_prediction/train_allele_specific_models_command.py b/mhcflurry/class1_affinity_prediction/train_allele_specific_models_command.py
index e8a67737eb989c0a6df1142abd7a9d0a40703e71..4b4201915e8700b6fe633f94db3d2b711edd0d5c 100644
--- a/mhcflurry/class1_affinity_prediction/train_allele_specific_models_command.py
+++ b/mhcflurry/class1_affinity_prediction/train_allele_specific_models_command.py
@@ -13,7 +13,7 @@ import pandas
import mhcnames
-from .class1_binding_predictor import Class1BindingPredictor
+from .class1_neural_network import Class1NeuralNetwork
from ..common import configure_logging
def normalize_allele_name(s):
@@ -122,7 +122,7 @@ def run():
train_data = df.ix[df.allele == allele].dropna().sample(
frac=1.0)
- model = Class1BindingPredictor(
+ model = Class1NeuralNetwork(
verbose=args.verbosity,
**hyperparameters)
diff --git a/mhcflurry/predict_command.py b/mhcflurry/predict_command.py
index 2231c2a2f34f1cdba74acd6d92432b9567f1d811..58c37c0c21bd48306013d6108ed23922495c888e 100644
--- a/mhcflurry/predict_command.py
+++ b/mhcflurry/predict_command.py
@@ -46,7 +46,8 @@ import pandas
import itertools
from .downloads import get_path
-from . import class1_affinity_prediction, class1_allele_specific_ensemble
+from .class1_affinity_prediction import Class1AffinityPredictor
+
parser = argparse.ArgumentParser(
description=__doc__,
@@ -94,24 +95,11 @@ parser.add_argument(
help="Output column name for predictions. Default: '%(default)s'")
parser.add_argument(
- "--predictor",
- choices=("class1-allele-specific-single", "class1-allele-specific-ensemble"),
- default="class1-allele-specific-ensemble",
- help="Predictor to use. Default: %(default)s.")
-
-parser.add_argument(
- "--models-class1-allele-specific-ensemble",
- metavar="DIR",
- default=None,
- help="Directory containing class1 allele specific ensemble models. "
- "Default: %s" % get_path("models_class1_allele_specific_ensemble", test_exists=False))
-
-parser.add_argument(
- "--models-class1-allele-specific-single",
+ "--models",
metavar="DIR",
default=None,
- help="Directory containing class1 allele specific single models. "
- "Default: %s" % get_path("models_class1_allele_specific_single", test_exists=False))
+ help="Directory containing models. "
+ "Default: %s" % get_path("models_class1", test_exists=False))
def run(argv=sys.argv[1:]):
@@ -155,29 +143,13 @@ def run(argv=sys.argv[1:]):
print("Predicting for %d alleles and %d peptides = %d predictions" % (
len(args.alleles), len(args.peptides), len(df)))
- if args.predictor == "class1-allele-specific-single":
- models_dir = args.models_class1_allele_specific_single
- if models_dir is None:
- # The reason we set the default here instead of in the argument parser is that
- # we want to test_exists at this point, so the user gets a message instructing
- # them to download the models if needed.
- models_dir = get_path("models_class1_allele_specific_single")
- predictor = (
- class1_affinity_prediction
- .class1_single_model_multi_allele_predictor
- .Class1SingleModelMultiAllelePredictor
- ).load_from_download_directory(models_dir)
- elif args.predictor == "class1-allele-specific-ensemble":
- models_dir = args.models_class1_allele_specific_ensemble
- if models_dir is None:
- models_dir = get_path("models_class1_allele_specific_ensemble")
- predictor = (
- class1_allele_specific_ensemble
- .class1_ensemble_multi_allele_predictor
- .Class1EnsembleMultiAllelePredictor
- ).load_from_download_directory(models_dir)
- else:
- assert False
+ models_dir = args.models
+ if models_dir is None:
+ # The reason we set the default here instead of in the argument parser is that
+ # we want to test_exists at this point, so the user gets a message instructing
+ # them to download the models if needed.
+ models_dir = get_path("models_class1")
+ predictor = Class1AffinityPredictor.load(models_dir)
predictions = {} # allele -> peptide -> value
for (allele, sub_df) in df.groupby(args.allele_column):
diff --git a/test/dummy_models.py b/test/dummy_models.py
index 0ea99cca80b5d8e71cb27e93c88f69d08fe6b6f1..5fb9facc3bcf6c9dc0d109201a7a222c93c5ebf4 100644
--- a/test/dummy_models.py
+++ b/test/dummy_models.py
@@ -13,7 +13,7 @@
# limitations under the License.
import numpy as np
-from mhcflurry import Class1BindingPredictor
+from mhcflurry import Class1NeuralNetwork
class Dummy9merIndexEncodingModel(object):
"""
@@ -30,19 +30,19 @@ class Dummy9merIndexEncodingModel(object):
n_cols,)
return np.ones(n_rows, dtype=float) * self.constant_output_value
-always_zero_predictor_with_unknown_AAs = Class1BindingPredictor(
+always_zero_predictor_with_unknown_AAs = Class1NeuralNetwork(
model=Dummy9merIndexEncodingModel(0),
allow_unknown_amino_acids=True)
-always_zero_predictor_without_unknown_AAs = Class1BindingPredictor(
+always_zero_predictor_without_unknown_AAs = Class1NeuralNetwork(
model=Dummy9merIndexEncodingModel(0),
allow_unknown_amino_acids=False)
-always_one_predictor_with_unknown_AAs = Class1BindingPredictor(
+always_one_predictor_with_unknown_AAs = Class1NeuralNetwork(
model=Dummy9merIndexEncodingModel(1),
allow_unknown_amino_acids=True)
-always_one_predictor_without_unknown_AAs = Class1BindingPredictor(
+always_one_predictor_without_unknown_AAs = Class1NeuralNetwork(
model=Dummy9merIndexEncodingModel(1),
allow_unknown_amino_acids=False)
diff --git a/test/test_class1_binding_predictor_A0205.py b/test/test_class1_binding_predictor_A0205.py
index 451d3ca734cb860655edf04fe4577eae54ef4a7e..9010e99a703f0a846321e19c4a33164fba631df5 100644
--- a/test/test_class1_binding_predictor_A0205.py
+++ b/test/test_class1_binding_predictor_A0205.py
@@ -2,7 +2,7 @@ import numpy
import pandas
numpy.random.seed(0)
-from mhcflurry import Class1BindingPredictor
+from mhcflurry import Class1NeuralNetwork
from nose.tools import eq_
from numpy import testing
@@ -25,7 +25,7 @@ def test_class1_binding_predictor_A0205_training_accuracy():
df.measurement_source == "kim2014"
]
- predictor = Class1BindingPredictor(
+ predictor = Class1NeuralNetwork(
activation="tanh",
layer_sizes=[64],
max_epochs=1000, # Memorize the dataset.
diff --git a/test/test_hyperparameters.py b/test/test_hyperparameters.py
index a840bbf7b458e3c3c7fa3cb0f0eafdab9c834c57..acc0d417498c60cc01132a749abc030cadff180f 100644
--- a/test/test_hyperparameters.py
+++ b/test/test_hyperparameters.py
@@ -1,6 +1,6 @@
from numpy.testing import assert_equal
-from mhcflurry.class1_affinity_prediction import Class1BindingPredictor
+from mhcflurry.class1_affinity_prediction import Class1NeuralNetwork
def test_all_combinations_of_hyperparameters():
@@ -8,7 +8,7 @@ def test_all_combinations_of_hyperparameters():
activation=["tanh", "sigmoid"],
fraction_negative=[0, 0.2])
results = (
- Class1BindingPredictor
+ Class1NeuralNetwork
.hyperparameter_defaults
.models_grid(**combinations_dict))
assert_equal(len(results), 4)
diff --git a/test/test_imputation.py b/test/test_imputation.py
index d30f12f7229475ccbed003e6cda8ffef204450ec..8376801851d20670808c83dcd7a5e680d1dcb308 100644
--- a/test/test_imputation.py
+++ b/test/test_imputation.py
@@ -3,7 +3,7 @@ np.random.seed(0)
from mhcflurry.imputation_helpers import imputer_from_name
from mhcflurry.affinity_measurement_dataset import AffinityMeasurementDataset
-from mhcflurry import Class1BindingPredictor
+from mhcflurry import Class1NeuralNetwork
from fancyimpute import MICE, KNN, SoftImpute, IterativeSVD
from nose.tools import eq_
@@ -54,7 +54,7 @@ def test_performance_improves_for_A0205_with_pretraining():
print("AffinityMeasurementDataset with only A0205, # entries: %d" % (
len(a0205_data_without_imputation)))
- predictor_without_imputation = Class1BindingPredictor(
+ predictor_without_imputation = Class1NeuralNetwork(
name="A0205-no-impute")
X_index, ic50_true, sample_weights, _ = (
@@ -97,7 +97,7 @@ def test_performance_improves_for_A0205_with_pretraining():
assert sample_weights_imputed.max() <= 1, sample_weights_imputed.max()
assert ic50_imputed.min() >= 0, ic50_imputed.min()
- predictor_with_imputation = Class1BindingPredictor(name="A0205-impute")
+ predictor_with_imputation = Class1NeuralNetwork(name="A0205-impute")
predictor_with_imputation.fit_kmer_encoded_arrays(
X=X_index,
diff --git a/test/test_serialization.py b/test/test_serialization.py
index 98870bb89eb0b66fb565aa06f039757bfe3f7f1a..e06ffc2cc366be971e906a5111666698cfedb529 100644
--- a/test/test_serialization.py
+++ b/test/test_serialization.py
@@ -1,13 +1,13 @@
import pickle
import numpy as np
-from mhcflurry.class1_affinity_prediction import Class1BindingPredictor
+from mhcflurry.class1_affinity_prediction import Class1NeuralNetwork
def test_predict_after_saving_model_to_disk():
# don't even bother fitting the model, just save its random weights
# and check we get the same predictions back afterward
- model = Class1BindingPredictor(name="rando")
+ model = Class1NeuralNetwork(name="rando")
peptides = ["A" * 9, "C" * 9]
original_predictions = model.predict(peptides)