From b09ed3b403785cffe9dfc4bdada6ec8c6fb3a115 Mon Sep 17 00:00:00 2001
From: Tim O'Donnell <timodonnell@gmail.com>
Date: Mon, 22 Apr 2019 13:46:06 -0400
Subject: [PATCH] cleanup hyperparameters
---
mhcflurry/class1_neural_network.py | 92 +++++++++++-------------------
mhcflurry/encodable_sequences.py | 52 +++++++++++------
2 files changed, 69 insertions(+), 75 deletions(-)
diff --git a/mhcflurry/class1_neural_network.py b/mhcflurry/class1_neural_network.py
index f47a3eeb..a9c1edb5 100644
--- a/mhcflurry/class1_neural_network.py
+++ b/mhcflurry/class1_neural_network.py
@@ -9,7 +9,7 @@ import pandas
from .hyperparameters import HyperparameterDefaults
-from .encodable_sequences import EncodableSequences
+from .encodable_sequences import EncodableSequences, EncodingError
from .amino_acid import available_vector_encodings, vector_encoding_length
from .regression_target import to_ic50, from_ic50
from .common import random_peptides, amino_acid_distribution
@@ -28,12 +28,15 @@ class Class1NeuralNetwork(object):
"""
network_hyperparameter_defaults = HyperparameterDefaults(
- kmer_size=15,
- peptide_amino_acid_encoding="BLOSUM62",
allele_amino_acid_encoding="BLOSUM62",
- embedding_input_dim=21,
- embedding_output_dim=8,
allele_dense_layer_sizes=[],
+ peptide_encoding={
+ 'vector_encoding_name': 'BLOSUM62',
+ 'alignment_method': 'pad_middle',
+ 'left_edge': 4,
+ 'right_edge': 4,
+ 'max_length': 15,
+ },
peptide_dense_layer_sizes=[],
peptide_allele_merge_method="multiply",
peptide_allele_merge_activation="",
@@ -45,7 +48,6 @@ class Class1NeuralNetwork(object):
output_activation="sigmoid",
dropout_probability=0.0,
batch_normalization=False,
- embedding_init_method="glorot_uniform",
locally_connected_layers=[
{
"filters": 8,
@@ -69,15 +71,6 @@ class Class1NeuralNetwork(object):
used.
"""
- input_encoding_hyperparameter_defaults = HyperparameterDefaults(
- alignment_method="pad_middle",
- left_edge=4,
- right_edge=4)
- """
- Number of amino acid residues that are given fixed positions on the each
- side in the variable length encoding.
- """
-
fit_hyperparameter_defaults = HyperparameterDefaults(
max_epochs=500,
validation_split=0.1,
@@ -110,7 +103,6 @@ class Class1NeuralNetwork(object):
hyperparameter_defaults = network_hyperparameter_defaults.extend(
compile_hyperparameter_defaults).extend(
- input_encoding_hyperparameter_defaults).extend(
fit_hyperparameter_defaults).extend(
early_stopping_hyperparameter_defaults).extend(
miscelaneous_hyperparameter_defaults
@@ -132,6 +124,13 @@ class Class1NeuralNetwork(object):
"verbose": None,
"mode": None,
"take_best_epoch": None,
+ 'kmer_size': None,
+ 'peptide_amino_acid_encoding': None,
+ 'embedding_input_dim': None,
+ 'embedding_output_dim': None,
+ 'embedding_init_method': None,
+ 'left_edge': None,
+ 'right_edge': None,
}
@classmethod
@@ -375,22 +374,8 @@ class Class1NeuralNetwork(object):
numpy.array
"""
encoder = EncodableSequences.create(peptides)
- if (self.hyperparameters['peptide_amino_acid_encoding'] == "embedding"):
- encoded = encoder.variable_length_to_fixed_length_categorical(
- max_length=self.hyperparameters['kmer_size'],
- **self.input_encoding_hyperparameter_defaults.subselect(
- self.hyperparameters))
- elif (
- self.hyperparameters['peptide_amino_acid_encoding'] in
- available_vector_encodings()):
- encoded = encoder.variable_length_to_fixed_length_vector_encoding(
- self.hyperparameters['peptide_amino_acid_encoding'],
- max_length=self.hyperparameters['kmer_size'],
- **self.input_encoding_hyperparameter_defaults.subselect(
- self.hyperparameters))
- else:
- raise ValueError("Unsupported peptide_amino_acid_encoding: %s" %
- self.hyperparameters['peptide_amino_acid_encoding'])
+ encoded = encoder.variable_length_to_fixed_length_vector_encoding(
+ **self.hyperparameters['peptide_encoding'])
assert len(encoded) == len(peptides)
return encoded
@@ -404,10 +389,16 @@ class Class1NeuralNetwork(object):
(int, int) tuple
"""
- return (
- self.hyperparameters['left_edge'] +
- self.hyperparameters['right_edge'],
- self.hyperparameters['kmer_size'])
+ # We currently have an arbitrary hard floor of 5, even if the underlying
+ # peptide encoding supports smaller lengths.
+ #
+ # We empirically find the supported peptide lengths based on the
+ # lengths for which peptides_to_network_input throws ValueError.
+ try:
+ self.peptides_to_network_input([""])
+ except EncodingError as e:
+ return e.supported_peptide_lengths
+ raise RuntimeError("peptides_to_network_input did not raise")
def allele_encoding_to_network_input(self, allele_encoding):
"""
@@ -799,11 +790,8 @@ class Class1NeuralNetwork(object):
def make_network(
self,
- kmer_size,
+ peptide_encoding,
allele_amino_acid_encoding,
- peptide_amino_acid_encoding,
- embedding_input_dim,
- embedding_output_dim,
allele_dense_layer_sizes,
peptide_dense_layer_sizes,
peptide_allele_merge_method,
@@ -816,7 +804,6 @@ class Class1NeuralNetwork(object):
output_activation,
dropout_probability,
batch_normalization,
- embedding_init_method,
locally_connected_layers,
allele_representations=None):
"""
@@ -832,23 +819,12 @@ class Class1NeuralNetwork(object):
from keras.layers.embeddings import Embedding
from keras.layers.normalization import BatchNormalization
- if peptide_amino_acid_encoding == "embedding":
- peptide_input = Input(
- shape=(kmer_size,), dtype='int32', name='peptide')
- current_layer = Embedding(
- input_dim=embedding_input_dim,
- output_dim=embedding_output_dim,
- input_length=kmer_size,
- embeddings_initializer=embedding_init_method,
- name="peptide_embedding")(peptide_input)
- else:
- peptide_input = Input(
- shape=(
- kmer_size,
- vector_encoding_length(peptide_amino_acid_encoding)),
- dtype='float32',
- name='peptide')
- current_layer = peptide_input
+ peptide_encoding_shape = self.peptides_to_network_input([]).shape[1:]
+ peptide_input = Input(
+ shape=peptide_encoding_shape,
+ dtype='float32',
+ name='peptide')
+ current_layer = peptide_input
inputs = [peptide_input]
diff --git a/mhcflurry/encodable_sequences.py b/mhcflurry/encodable_sequences.py
index 84e9a2da..0f0fcbd9 100644
--- a/mhcflurry/encodable_sequences.py
+++ b/mhcflurry/encodable_sequences.py
@@ -13,6 +13,14 @@ import pandas
from . import amino_acid
+class EncodingError(ValueError):
+ def __init__(self, message, supported_peptide_lengths):
+ self.supported_peptide_lengths = supported_peptide_lengths
+ ValueError.__init__(
+ self,
+ message + " Supported lengths: %s - %s." % supported_peptide_lengths)
+
+
class EncodableSequences(object):
"""
Sequences of amino acids.
@@ -36,7 +44,7 @@ class EncodableSequences(object):
if not all(isinstance(obj, string_types) for obj in sequences):
raise ValueError("Sequence of strings is required")
self.sequences = numpy.array(sequences)
- lengths = pandas.Series(self.sequences).str.len()
+ lengths = pandas.Series(self.sequences, dtype=numpy.object_).str.len()
self.min_length = lengths.min()
self.max_length = lengths.max()
@@ -187,26 +195,23 @@ class EncodableSequences(object):
shape=(len(sequences), max_length),
dtype="int32")
- df = pandas.DataFrame({"peptide": sequences})
+ df = pandas.DataFrame({"peptide": sequences}, dtype=numpy.object_)
df["length"] = df.peptide.str.len()
middle_length = max_length - left_edge - right_edge
+ min_length = left_edge + right_edge
# For efficiency we handle each supported peptide length using bulk
# array operations.
for (length, sub_df) in df.groupby("length"):
- if length < left_edge + right_edge:
- raise ValueError(
- "Sequence '%s' (length %d) unsupported: length must be at "
- "least %d. There are %d total peptides with this length." % (
- sub_df.iloc[0].peptide, length, left_edge + right_edge,
- len(sub_df)))
- if length > max_length:
- raise ValueError(
- "Sequence '%s' (length %d) unsupported: length must be at "
- "most %d. There are %d total peptides with this length." % (
- sub_df.iloc[0].peptide, length, max_length,
- len(sub_df)))
+ if length < min_length or length > max_length:
+ raise EncodingError(
+ "Sequence '%s' (length %d) unsupported. There are %d "
+ "total peptides with this length." % (
+ sub_df.iloc[0].peptide,
+ length,
+ len(sub_df)), supported_peptide_lengths=(
+ min_length, max_length))
# Array of shape (num peptides, length) giving fixed-length amino
# acid encoding each peptide of the current length.
@@ -240,17 +245,30 @@ class EncodableSequences(object):
-right_edge:
] = fixed_length_sequences[:, -right_edge:]
elif alignment_method == "left_pad_right_pad":
+ # We arbitrarily set a minimum length of 5, although this encoding
+ # could handle smaller peptides.
+ min_length = 5
+
# Result array is int32, filled with X (null amino acid) value.
result = numpy.full(fill_value=amino_acid.AMINO_ACID_INDEX['X'],
shape=(len(sequences), max_length * 2), dtype="int32")
- df = pandas.DataFrame({"peptide": sequences})
+ df = pandas.DataFrame({"peptide": sequences}, dtype=numpy.object_)
# For efficiency we handle each supported peptide length using bulk
# array operations.
for (length, sub_df) in df.groupby(df.peptide.str.len()):
- # Array of shape (num peptides, length) giving fixed-length amino
- # acid encoding each peptide of the current length.
+ if length < min_length or length > max_length:
+ raise EncodingError(
+ "Sequence '%s' (length %d) unsupported. There are %d "
+ "total peptides with this length." % (
+ sub_df.iloc[0].peptide,
+ length,
+ len(sub_df)), supported_peptide_lengths=(
+ min_length, max_length))
+
+ # Array of shape (num peptides, length) giving fixed-length
+ # amino acid encoding each peptide of the current length.
fixed_length_sequences = numpy.stack(sub_df.peptide.map(
lambda s: numpy.array(
[amino_acid.AMINO_ACID_INDEX[char] for char in
--
GitLab