Newer
Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
from __future__ import print_function
from os.path import join, exists, abspath
from os import mkdir
from socket import gethostname
from getpass import getuser
import time
import json
import hashlib
import logging
from six import string_types
import numpy
import pandas
from .version import __version__
from .class1_neural_network import DEFAULT_PREDICT_BATCH_SIZE
from .encodable_sequences import EncodableSequences
from .downloads import get_default_class1_cleavage_models_dir
from .class1_cleavage_neural_network import Class1CleavageNeuralNetwork
from .common import save_weights, load_weights, NumpyJSONEncoder
class Class1CleavagePredictor(object):
def __init__(
self,
models,
manifest_df=None,
metadata_dataframes=None):
self.models = models
self._manifest_df = manifest_df
self.metadata_dataframes = (
dict(metadata_dataframes) if metadata_dataframes else {})
@property
def manifest_df(self):
"""
A pandas.DataFrame describing the models included in this predictor.
Returns
-------
pandas.DataFrame
"""
if self._manifest_df is None:
rows = []
for (i, model) in enumerate(self.models):
model_config = model.get_config()
rows.append((
self.model_name(i),
json.dumps(model_config, cls=NumpyJSONEncoder),
model
))
self._manifest_df = pandas.DataFrame(
rows,
columns=["model_name", "config_json", "model"])
return self._manifest_df
@staticmethod
def model_name(num):
"""
Generate a model name
Returns
-------
string
"""
random_string = hashlib.sha1(
str(time.time()).encode()).hexdigest()[:16]
return "CLEAVAGE-CLASSI-%d-%s" % (
num,
random_string)
@staticmethod
def weights_path(models_dir, model_name):
"""
Generate the path to the weights file for a model
Parameters
----------
models_dir : string
model_name : string
Returns
-------
string
"""
return join(models_dir, "weights_%s.npz" % model_name)
def predict(
self,
peptides,
n_flanks,
c_flanks,
batch_size=DEFAULT_PREDICT_BATCH_SIZE):
return self.predict_to_dataframe(
peptides=peptides,
n_flanks=n_flanks,
c_flanks=c_flanks,
batch_size=batch_size).score.values
def predict_to_dataframe(
self,
peptides,
n_flanks,
c_flanks,
batch_size=DEFAULT_PREDICT_BATCH_SIZE):
for (name, value) in [
("peptides", peptides),
("n_flanks", n_flanks),
("c_flanks", c_flanks)]:
if isinstance(value, string_types):
raise TypeError(
"%s must be a list or array, not a string" % name)
peptides = EncodableSequences.create(peptides)
n_flanks = EncodableSequences.create(n_flanks)
c_flanks = EncodableSequences.create(c_flanks)
score_array = []
for (i, network) in enumerate(self.models):
predictions = network.predict(
peptides=peptides,
n_flanks=n_flanks,
c_flanks=c_flanks,
batch_size=batch_size)
score_array.append(predictions)
score_array = numpy.array(score_array)
result_df = pandas.DataFrame({
"peptide": peptides.sequences,
"n_flank": n_flanks.sequences,
"c_flank": c_flanks.sequences,
"score": numpy.mean(score_array, axis=0),
})
return result_df
def check_consistency(self):
"""
Verify that self.manifest_df is consistent with instance variables.
Currently only checks for agreement on the total number of models.
Throws AssertionError if inconsistent.
"""
assert len(self.manifest_df) == len(self.models), (
"Manifest seems out of sync with models: %d vs %d entries: \n%s"% (
len(self.manifest_df),
len(self.models),
str(self.manifest_df)))
def save(self, models_dir, model_names_to_write=None, write_metadata=True):
"""
Serialize the predictor to a directory on disk. If the directory does
not exist it will be created.
The serialization format consists of a file called "manifest.csv" with
the configurations of each Class1CleavageNeuralNetwork, along with
per-network files giving the model weights.
Parameters
----------
models_dir : string
Path to directory. It will be created if it doesn't exist.
"""
self.check_consistency()
if model_names_to_write is None:
# Write all models
model_names_to_write = self.manifest_df.model_name.values
if not exists(models_dir):
mkdir(models_dir)
sub_manifest_df = self.manifest_df.loc[
self.manifest_df.model_name.isin(model_names_to_write)
].copy()
# Network JSON configs may have changed since the models were added,
# so we update the JSON configs here also.
updated_network_config_jsons = []
for (_, row) in sub_manifest_df.iterrows():
updated_network_config_jsons.append(
json.dumps(row.model.get_config(), cls=NumpyJSONEncoder))
weights_path = self.weights_path(models_dir, row.model_name)
save_weights(row.model.get_weights(), weights_path)
logging.info("Wrote: %s", weights_path)
sub_manifest_df["config_json"] = updated_network_config_jsons
self.manifest_df.loc[
sub_manifest_df.index,
"config_json"
] = updated_network_config_jsons
write_manifest_df = self.manifest_df[[
c for c in self.manifest_df.columns if c != "model"
]]
manifest_path = join(models_dir, "manifest.csv")
write_manifest_df.to_csv(manifest_path, index=False)
logging.info("Wrote: %s", manifest_path)
if write_metadata:
# Write "info.txt"
info_path = join(models_dir, "info.txt")
rows = [
("trained on", time.asctime()),
("package ", "mhcflurry %s" % __version__),
("hostname ", gethostname()),
("user ", getuser()),
]
pandas.DataFrame(rows).to_csv(
info_path, sep="\t", header=False, index=False)
if self.metadata_dataframes:
for (name, df) in self.metadata_dataframes.items():
metadata_df_path = join(models_dir, "%s.csv.bz2" % name)
df.to_csv(metadata_df_path, index=False, compression="bz2")
@classmethod
def load(cls, models_dir=None, max_models=None):
"""
Deserialize a predictor from a directory on disk.
Parameters
----------
models_dir : string
Path to directory. If unspecified the default downloaded models are
used.
max_models : int, optional
Maximum number of models to load
Returns
-------
`Class1CleavagePredictor` instance
"""
if models_dir is None:
models_dir = get_default_class1_cleavage_models_dir()
manifest_path = join(models_dir, "manifest.csv")
manifest_df = pandas.read_csv(manifest_path, nrows=max_models)
models = []
for (_, row) in manifest_df.iterrows():
weights_filename = cls.weights_path(models_dir, row.model_name)
config = json.loads(row.config_json)
model = Class1CleavageNeuralNetwork.from_config(
config,
weights=load_weights(abspath(weights_filename)))
models.append(model)
manifest_df["model"] = models
logging.info("Loaded %d class1 cleavage models", len(models))
result = cls(
models=models,
manifest_df=manifest_df)
return result