Merge pull request #82 from hartym/feature/io_pickle

Feature/io pickle
This commit is contained in:
Romain Dorgueil
2017-05-25 07:47:58 -07:00
committed by GitHub
6 changed files with 193 additions and 2 deletions

View File

@ -1,6 +1,6 @@
from bonobo.structs import Bag, Graph, Token
from bonobo.nodes import CsvReader, CsvWriter, FileReader, FileWriter, Filter, JsonReader, JsonWriter, Limit, \
PrettyPrint, Tee, count, identity, noop, pprint
PrettyPrint, PickleWriter, PickleReader, Tee, count, identity, noop, pprint
from bonobo.strategies import create_strategy
from bonobo.util.objects import get_name
@ -43,7 +43,6 @@ def run(graph, strategy=None, plugins=None, services=None):
plugins = plugins or []
from bonobo import settings
settings.check()
if not settings.QUIET: # pragma: no cover
@ -98,6 +97,8 @@ register_api_group(
JsonReader,
JsonWriter,
Limit,
PickleReader,
PickleWriter,
PrettyPrint,
Tee,
count,

Binary file not shown.

View File

@ -0,0 +1,58 @@
import bonobo
from fs.tarfs import TarFS
import os
def cleanse_sms(row):
if row['category'] == 'spam':
row['sms_clean'] = '**MARKED AS SPAM** ' + row['sms'][0:50] + ('...' if len(row['sms']) > 50 else '')
else:
row['sms_clean'] = row['sms']
return row['sms_clean']
graph = bonobo.Graph(
bonobo.PickleReader('spam.pkl'), # spam.pkl is within the gzipped tarball
cleanse_sms,
print
)
if __name__ == '__main__':
'''
This example shows how a different file system service can be injected
into a transformation (as compressing pickled objects often makes sense
anyways). The pickle itself contains a list of lists as follows:
```
[
['category', 'sms'],
['ham', 'Go until jurong point, crazy..'],
['ham', 'Ok lar... Joking wif u oni...'],
['spam', 'Free entry in 2 a wkly comp to win...'],
['ham', 'U dun say so early hor... U c already then say...'],
['ham', 'Nah I don't think he goes to usf, he lives around here though'],
['spam', 'FreeMsg Hey there darling it's been 3 week's now...'],
...
]
```
where the first column categorizes and sms as "ham" or "spam". The second
column contains the sms itself.
Data set taken from:
https://www.kaggle.com/uciml/sms-spam-collection-dataset/downloads/sms-spam-collection-dataset.zip
The transformation (1) reads the pickled data, (2) marks and shortens
messages categorized as spam, and (3) prints the output.
'''
services = {
'fs': TarFS(
os.path.join(bonobo.get_examples_path(), 'datasets', 'spam.tgz')
)
}
bonobo.run(graph, services=services)

View File

@ -3,6 +3,7 @@
from .file import FileReader, FileWriter
from .json import JsonReader, JsonWriter
from .csv import CsvReader, CsvWriter
from .pickle import PickleReader, PickleWriter
__all__ = [
'CsvReader',
@ -11,4 +12,6 @@ __all__ = [
'FileWriter',
'JsonReader',
'JsonWriter',
'PickleReader',
'PickleWriter',
]

69
bonobo/nodes/io/pickle.py Normal file
View File

@ -0,0 +1,69 @@
import pickle
from bonobo.config.processors import ContextProcessor
from bonobo.config import Option
from bonobo.constants import NOT_MODIFIED
from bonobo.util.objects import ValueHolder
from .file import FileReader, FileWriter, FileHandler
class PickleHandler(FileHandler):
"""
.. attribute:: item_names
The names of the items in the pickle, if it is not defined in the first item of the pickle.
"""
item_names = Option(tuple)
class PickleReader(PickleHandler, FileReader):
"""
Reads a Python pickle object and yields the items in dicts.
"""
mode = Option(str, default='rb')
@ContextProcessor
def pickle_headers(self, context, fs, file):
yield ValueHolder(self.item_names)
def read(self, fs, file, pickle_headers):
data = pickle.load(file)
# if the data is not iterable, then wrap the object in a list so it may be iterated
if isinstance(data, dict):
is_dict = True
iterator = iter(data.items())
else:
is_dict = False
try:
iterator = iter(data)
except TypeError:
iterator = iter([data])
if not pickle_headers.get():
pickle_headers.set(next(iterator))
item_count = len(pickle_headers.value)
for i in iterator:
if len(i) != item_count:
raise ValueError('Received an object with %d items, expecting %d.' % (len(i), item_count, ))
yield dict(zip(i)) if is_dict else dict(zip(pickle_headers.value, i))
class PickleWriter(PickleHandler, FileWriter):
mode = Option(str, default='wb')
def write(self, fs, file, lineno, item):
"""
Write a pickled item to the opened file.
"""
file.write(pickle.dumps(item))
lineno += 1
return NOT_MODIFIED

60
tests/io/test_pickle.py Normal file
View File

@ -0,0 +1,60 @@
import pickle
import pytest
from bonobo import Bag, PickleReader, PickleWriter, open_fs
from bonobo.constants import BEGIN, END
from bonobo.execution.node import NodeExecutionContext
from bonobo.util.testing import CapturingNodeExecutionContext
def test_write_pickled_dict_to_file(tmpdir):
fs, filename = open_fs(tmpdir), 'output.pkl'
writer = PickleWriter(path=filename)
context = NodeExecutionContext(writer, services={'fs': fs})
context.write(BEGIN, Bag({'foo': 'bar'}), Bag({'foo': 'baz', 'ignore': 'this'}), END)
context.start()
context.step()
context.step()
context.stop()
assert pickle.loads(fs.open(filename, 'rb').read()) == {'foo': 'bar'}
with pytest.raises(AttributeError):
getattr(context, 'file')
def test_read_pickled_list_from_file(tmpdir):
fs, filename = open_fs(tmpdir), 'input.pkl'
fs.open(filename, 'wb').write(pickle.dumps([
['a', 'b', 'c'], ['a foo', 'b foo', 'c foo'], ['a bar', 'b bar', 'c bar']
]))
reader = PickleReader(path=filename)
context = CapturingNodeExecutionContext(reader, services={'fs': fs})
context.start()
context.write(BEGIN, Bag(), END)
context.step()
context.stop()
assert len(context.send.mock_calls) == 2
args0, kwargs0 = context.send.call_args_list[0]
assert len(args0) == 1 and not len(kwargs0)
args1, kwargs1 = context.send.call_args_list[1]
assert len(args1) == 1 and not len(kwargs1)
assert args0[0].args[0] == {
'a': 'a foo',
'b': 'b foo',
'c': 'c foo',
}
assert args1[0].args[0] == {
'a': 'a bar',
'b': 'b bar',
'c': 'c bar',
}