From ac3f874fa90dc032bf592b2f364a5564c9cd8373 Mon Sep 17 00:00:00 2001
From: Tim O'Donnell <timodonnell@gmail.com>
Date: Sat, 25 Nov 2017 11:13:12 -0500
Subject: [PATCH] Fix tests, add logging to calibrate_percentile_ranks, add
SIGUSR1 trick to train command
---
.../class1_affinity_predictor.py | 27 ++++++++++++++++---
.../class1_neural_network.py | 2 +-
.../train_allele_specific_models_command.py | 6 +++++
test/test_class1_affinity_predictor.py | 12 +++------
4 files changed, 35 insertions(+), 12 deletions(-)
diff --git a/mhcflurry/class1_affinity_prediction/class1_affinity_predictor.py b/mhcflurry/class1_affinity_prediction/class1_affinity_predictor.py
index c0ec13fb..e61d46f7 100644
--- a/mhcflurry/class1_affinity_prediction/class1_affinity_predictor.py
+++ b/mhcflurry/class1_affinity_prediction/class1_affinity_predictor.py
@@ -484,9 +484,10 @@ class Class1AffinityPredictor(object):
def calibrate_percentile_ranks(
self,
peptides=None,
- num_peptides_per_length=int(1e6),
+ num_peptides_per_length=int(1e5),
alleles=None,
- bins=None):
+ bins=None,
+ quiet=False):
"""
Compute the cumulative distribution of ic50 values for a set of alleles
over a large universe of random peptides, to enable computing quantiles in
@@ -503,6 +504,12 @@ class Class1AffinityPredictor(object):
alleles : sequence of string, optional
Alleles to perform calibration for. If not specified all supported
alleles will be calibrated.
+ bins : object
+ Anything that can be passed to numpy.histogram's "bins" argument
+ can be used here, i.e. either an integer or a sequence giving bin
+ edges. This is in ic50 space.
+ quiet : boolean
+ If False (default), status updates will be printed to stdout.
"""
if bins is None:
bins = to_ic50(numpy.linspace(1, 0, 1000))
@@ -519,11 +526,25 @@ class Class1AffinityPredictor(object):
peptides.extend(
random_peptides(num_peptides_per_length, length))
- for allele in alleles:
+ if quiet:
+ def msg(s):
+ pass
+ else:
+ def msg(s):
+ print(s)
+
+ for (i, allele) in enumerate(alleles):
+ msg("Calibrating percentile ranks for allele %03d/%03d: %s" % (
+ i + 1, len(alleles), allele))
+ start = time.time()
predictions = self.predict(peptides, allele=allele)
+ msg("Generated %d predictions in %0.2f sec." % (
+ len(predictions), time.time() - start))
transform = PercentRankTransform()
transform.fit(predictions, bins=bins)
self.allele_to_percent_rank_transform[allele] = transform
+ msg("Done calibrating allele %s in %0.2f sec." % (
+ allele, time.time() - start))
def percentile_ranks(self, affinities, allele=None, alleles=None, throw=True):
"""
diff --git a/mhcflurry/class1_affinity_prediction/class1_neural_network.py b/mhcflurry/class1_affinity_prediction/class1_neural_network.py
index d685669c..f37827f3 100644
--- a/mhcflurry/class1_affinity_prediction/class1_neural_network.py
+++ b/mhcflurry/class1_affinity_prediction/class1_neural_network.py
@@ -455,7 +455,7 @@ class Class1NeuralNetwork(object):
for (key, value) in fit_history.history.items():
self.loss_history[key].extend(value)
- logging.info(
+ print(
"Epoch %3d / %3d: loss=%g. Min val loss at epoch %s" % (
i,
self.hyperparameters['max_epochs'],
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 aa12add4..bdc90dd8 100644
--- a/mhcflurry/class1_affinity_prediction/train_allele_specific_models_command.py
+++ b/mhcflurry/class1_affinity_prediction/train_allele_specific_models_command.py
@@ -7,6 +7,8 @@ import sys
import argparse
import yaml
import time
+import signal
+import traceback
import pandas
@@ -65,6 +67,10 @@ parser.add_argument(
def run(argv=sys.argv[1:]):
+ # On sigusr1 print stack trace
+ print("To show stack trace, run:\nkill -s USR1 %d" % os.getpid())
+ signal.signal(signal.SIGUSR1, lambda sig, frame: traceback.print_stack())
+
args = parser.parse_args(argv)
configure_logging(verbose=args.verbosity > 1)
diff --git a/test/test_class1_affinity_predictor.py b/test/test_class1_affinity_predictor.py
index 952e4840..3ab6277c 100644
--- a/test/test_class1_affinity_predictor.py
+++ b/test/test_class1_affinity_predictor.py
@@ -59,7 +59,7 @@ def test_a1_known_epitopes_in_newly_trained_model():
]
hyperparameters = {
- "max_epochs": 500,
+ "max_epochs": 100,
"patience": 10,
"early_stopping": True,
"validation_split": 0.2,
@@ -67,15 +67,11 @@ def test_a1_known_epitopes_in_newly_trained_model():
"random_negative_rate": 0.0,
"random_negative_constant": 25,
+ "peptide_amino_acid_encoding": "BLOSUM62",
"use_embedding": False,
"kmer_size": 15,
"batch_normalization": False,
"locally_connected_layers": [
- {
- "filters": 8,
- "activation": "tanh",
- "kernel_size": 3
- },
{
"filters": 8,
"activation": "tanh",
@@ -129,7 +125,7 @@ def test_class1_affinity_predictor_a0205_memorize_training_data():
hyperparameters = dict(
activation="tanh",
layer_sizes=[64],
- max_epochs=500,
+ max_epochs=100,
early_stopping=False,
validation_split=0.0,
locally_connected_layers=[],
@@ -163,7 +159,7 @@ def test_class1_affinity_predictor_a0205_memorize_training_data():
peptides=df.peptide.values,
affinities=df.measurement_value.values,
)
- predictor.calibrate_percentile_ranks()
+ predictor.calibrate_percentile_ranks(num_peptides_per_length=1000)
ic50_pred = predictor.predict(df.peptide.values, allele=allele)
ic50_true = df.measurement_value.values
eq_(len(ic50_pred), len(ic50_true))
--
GitLab