From c0fbebe1b4b95c0988fdbe874d1f48706a743040 Mon Sep 17 00:00:00 2001
From: Tim O'Donnell <timodonnell@gmail.com>
Date: Tue, 28 Nov 2017 14:26:09 -0500
Subject: [PATCH] Clearer output during training
---
.../class1_affinity_predictor.py | 26 ++++++++++++++-----
.../class1_neural_network.py | 19 +++++++++-----
.../train_allele_specific_models_command.py | 18 ++++++++-----
3 files changed, 44 insertions(+), 19 deletions(-)
diff --git a/mhcflurry/class1_affinity_prediction/class1_affinity_predictor.py b/mhcflurry/class1_affinity_prediction/class1_affinity_predictor.py
index fa0623fb..d1e0f02c 100644
--- a/mhcflurry/class1_affinity_prediction/class1_affinity_predictor.py
+++ b/mhcflurry/class1_affinity_prediction/class1_affinity_predictor.py
@@ -370,7 +370,8 @@ class Class1AffinityPredictor(object):
peptides,
affinities,
models_dir_for_save=None,
- verbose=1):
+ verbose=1,
+ progress_preamble=""):
"""
Fit one or more allele specific predictors for a single allele using a
single neural network architecture.
@@ -399,6 +400,9 @@ class Class1AffinityPredictor(object):
verbose : int
Keras verbosity
+ progress_preamble : string
+ Optional string of information to include in each progress update
+
Returns
-------
list of Class1NeuralNetwork
@@ -411,7 +415,8 @@ class Class1AffinityPredictor(object):
peptides=peptides,
affinities=affinities,
allele_pseudosequences=None,
- verbose=verbose)
+ verbose=verbose,
+ progress_preamble=progress_preamble)
if allele not in self.allele_to_allele_specific_models:
self.allele_to_allele_specific_models[allele] = []
@@ -442,7 +447,8 @@ class Class1AffinityPredictor(object):
peptides,
affinities,
models_dir_for_save=None,
- verbose=1):
+ verbose=1,
+ progress_preamble=""):
"""
Fit one or more pan-allele predictors using a single neural network
architecture.
@@ -472,6 +478,9 @@ class Class1AffinityPredictor(object):
verbose : int
Keras verbosity
+ progress_preamble : string
+ Optional string of information to include in each progress update
+
Returns
-------
list of Class1NeuralNetwork
@@ -486,7 +495,8 @@ class Class1AffinityPredictor(object):
peptides=peptides,
affinities=affinities,
allele_pseudosequences=allele_pseudosequences,
- verbose=verbose)
+ verbose=verbose,
+ progress_preamble=progress_preamble)
for (i, model) in enumerate(models):
model_name = self.model_name("pan-class1", i)
@@ -511,7 +521,8 @@ class Class1AffinityPredictor(object):
peptides,
affinities,
allele_pseudosequences,
- verbose=1):
+ verbose=1,
+ progress_preamble = ""):
"""
Private helper method
@@ -523,6 +534,8 @@ class Class1AffinityPredictor(object):
affinities : list of float
allele_pseudosequences : EncodableSequences or list of string
verbose : int
+ progress_preamble : string
+ Optional string of information to include in each progress update
Returns
-------
@@ -536,7 +549,8 @@ class Class1AffinityPredictor(object):
encodable_peptides,
affinities,
allele_pseudosequences=allele_pseudosequences,
- verbose=verbose)
+ verbose=verbose,
+ progress_preamble=progress_preamble)
yield model
def calibrate_percentile_ranks(
diff --git a/mhcflurry/class1_affinity_prediction/class1_neural_network.py b/mhcflurry/class1_affinity_prediction/class1_neural_network.py
index 995a629f..48e7d75d 100644
--- a/mhcflurry/class1_affinity_prediction/class1_neural_network.py
+++ b/mhcflurry/class1_affinity_prediction/class1_neural_network.py
@@ -317,7 +317,8 @@ class Class1NeuralNetwork(object):
affinities,
allele_pseudosequences=None,
sample_weights=None,
- verbose=1):
+ verbose=1,
+ progress_preamble=""):
"""
Fit the neural network.
@@ -338,6 +339,9 @@ class Class1NeuralNetwork(object):
verbose : int
Keras verbosity level
+
+ progress_preamble : string
+ Optional string of information to include in each progress update
"""
self.fit_num_points = len(peptides)
@@ -453,11 +457,14 @@ class Class1NeuralNetwork(object):
self.loss_history[key].extend(value)
print(
- "Epoch %3d / %3d: loss=%g. Min val loss at epoch %s" % (
- i,
- self.hyperparameters['max_epochs'],
- self.loss_history['loss'][-1],
- min_val_loss_iteration))
+ (
+ progress_preamble + " " +
+ "Epoch %3d / %3d: loss=%g. Min val loss (%s) at epoch %s" % (
+ i,
+ self.hyperparameters['max_epochs'],
+ self.loss_history['loss'][-1],
+ str(min_val_loss),
+ min_val_loss_iteration)).strip())
if self.hyperparameters['validation_split']:
val_loss = self.loss_history['val_loss'][-1]
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 f71147e0..cc9f82a6 100644
--- a/mhcflurry/class1_affinity_prediction/train_allele_specific_models_command.py
+++ b/mhcflurry/class1_affinity_prediction/train_allele_specific_models_command.py
@@ -79,7 +79,7 @@ parser.add_argument(
"--verbosity",
type=int,
help="Keras verbosity. Default: %(default)s",
- default=1)
+ default=0)
parser.add_argument(
"--parallelization-num-jobs",
default=1,
@@ -172,6 +172,7 @@ def run(argv=sys.argv[1:]):
'allele': allele,
'sub_df': sub_df,
'hyperparameters': hyperparameters,
+ 'verbose': args.verbosity,
'predictor': predictor if not worker_pool else None,
'save_to': args.out_models_dir if not worker_pool else None,
}
@@ -220,22 +221,23 @@ def process_work(
allele,
sub_df,
hyperparameters,
+ verbose,
predictor,
save_to):
if predictor is None:
predictor = Class1AffinityPredictor()
- print(
+ progress_preamble = (
"[%2d / %2d hyperparameters] "
- "[%2d / %2d replicates] "
- "[%4d / %4d alleles]: %s" % (
+ "[%4d / %4d alleles] "
+ "[%2d / %2d replicates]: %s " % (
hyperparameter_set_num + 1,
num_hyperparameter_sets,
- model_group + 1,
- n_models,
allele_num + 1,
n_alleles,
+ model_group + 1,
+ n_models,
allele))
train_data = sub_df.dropna().sample(frac=1.0)
@@ -245,7 +247,9 @@ def process_work(
allele=allele,
peptides=train_data.peptide.values,
affinities=train_data.measurement_value.values,
- models_dir_for_save=save_to)
+ models_dir_for_save=save_to,
+ progress_preamble=progress_preamble,
+ verbose=verbose)
return predictor
--
GitLab