From f19e2e0a8f16fdecd48e733b0f7dce08310a027b Mon Sep 17 00:00:00 2001
From: Tim O'Donnell <timodonnell@gmail.com>
Date: Wed, 14 Feb 2018 11:02:05 -0500
Subject: [PATCH] update hyperparameters
---
mhcflurry/class1_neural_network.py | 10 +++++-----
1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/mhcflurry/class1_neural_network.py b/mhcflurry/class1_neural_network.py
index 9b284122..6cf42fcc 100644
--- a/mhcflurry/class1_neural_network.py
+++ b/mhcflurry/class1_neural_network.py
@@ -29,7 +29,7 @@ class Class1NeuralNetwork(object):
network_hyperparameter_defaults = HyperparameterDefaults(
kmer_size=15,
- peptide_amino_acid_encoding="one-hot",
+ peptide_amino_acid_encoding="BLOSUM62",
embedding_input_dim=21,
embedding_output_dim=8,
allele_dense_layer_sizes=[],
@@ -39,7 +39,7 @@ class Class1NeuralNetwork(object):
layer_sizes=[32],
dense_layer_l1_regularization=0.001,
dense_layer_l2_regularization=0.0,
- activation="relu",
+ activation="tanh",
init="glorot_uniform",
output_activation="sigmoid",
dropout_probability=0.0,
@@ -59,7 +59,7 @@ class Class1NeuralNetwork(object):
"""
compile_hyperparameter_defaults = HyperparameterDefaults(
- loss="mse",
+ loss="custom:mse_with_inequalities",
optimizer="rmsprop",
)
"""
@@ -78,10 +78,10 @@ class Class1NeuralNetwork(object):
fit_hyperparameter_defaults = HyperparameterDefaults(
max_epochs=500,
take_best_epoch=False, # currently unused
- validation_split=0.2,
+ validation_split=0.1,
early_stopping=True,
minibatch_size=128,
- random_negative_rate=0.0,
+ random_negative_rate=0.2,
random_negative_constant=25,
random_negative_affinity_min=20000.0,
random_negative_affinity_max=50000.0,
--
GitLab