def pretty_print(filename, message_type='guess'):
'''
Print text/json/protobuf messages from a raw/gz/bz2 file.
'''
parts = os.path.basename(filename).split('.')
if parts[-1] in ['gz', 'bz2']:
parts.pop()
protocol = parts[-1]
if protocol == 'json':
data = json_load(filename)
print json.dumps(data, sort_keys=True, indent=4)
elif protocol == 'pb':
message = get_message(filename, message_type)
with open_compressed(filename) as f:
message.ParseFromString(f.read())
print message
elif protocol == 'pbs':
message = get_message(filename, message_type)
for string in protobuf_stream_load(filename):
message.ParseFromString(string)
print message
elif protocol == 'pickle':
data = pickle_load(filename)
print repr(data)
else:
with open_compressed(filename) as f:
for line in f:
print line,
def import_rows():
row = Row()
pos = row.diff.pos
neg = row.diff.neg
pos.observed.sparsity = ProductValue.Observed.SPARSE
neg.observed.sparsity = ProductValue.Observed.SPARSE
with open_compressed(RAW) as infile:
doc_count = int(infile.next())
word_count = int(infile.next())
observed_count = int(infile.next())
print 'Importing {} observations of {} words in {} documents'.format(
observed_count, word_count, doc_count)
with open_compressed(DIFFS, 'wb') as outfile:
current_doc = None
for line in infile:
doc, feature, count = line.split()
if doc != current_doc:
if current_doc is not None:
pos.observed.sparse.sort()
neg.observed.sparse.sort()
protobuf_stream_write(row.SerializeToString(), outfile)
print_dot(every=1000)
current_doc = doc
row.id = int(doc)
del pos.booleans[:]
del pos.observed.sparse[:]
del neg.booleans[:]
del neg.observed.sparse[:]
feature = int(feature) - 1
pos.observed.sparse.append(feature)
pos.booleans.append(True)
neg.observed.sparse.append(feature)
neg.booleans.append(False)
protobuf_stream_write(row.SerializeToString(), outfile)
def test_predict(root, rows_csv, encoding, **unused):
with tempdir(cleanup_on_error=CLEANUP_ON_ERROR):
with loom.preql.get_server(root, debug=True) as preql:
result_out = 'predictions_out.csv'
rows_in = os.listdir(rows_csv)[0]
rows_in = os.path.join(rows_csv, rows_in)
preql.predict(rows_in, COUNT, result_out, id_offset=True)
print 'DEBUG', open_compressed(rows_in).read()
print 'DEBUG', open_compressed(result_out).read()
_check_predictions(rows_in, result_out, encoding)
def test_predict(root, rows_csv, encoding, **unused):
with tempdir(cleanup_on_error=CLEANUP_ON_ERROR):
with loom.preql.get_server(root, debug=True) as preql:
result_out = 'predictions_out.csv'
rows_in = os.listdir(rows_csv)[0]
rows_in = os.path.join(rows_csv, rows_in)
preql.predict(rows_in, COUNT, result_out, id_offset=True)
print 'DEBUG', open_compressed(rows_in).read()
print 'DEBUG', open_compressed(result_out).read()
_check_predictions(rows_in, result_out, encoding)
def force_ascii(filename_in, filename_out=None, size=4096):
with ExitStack() as stack:
with_ = stack.enter_context
if filename_out is None:
filename_out = with_(loom.util.temp_copy(filename_in))
source = with_(open_compressed(filename_in, 'rb'))
destin = with_(open_compressed(filename_out, 'w'))
chunk = source.read(size)
while chunk:
destin.write(re_nonascii.sub('', chunk))
chunk = source.read(size)
def config_dump(config, filename):
config = deepcopy(config)
fill_in_defaults(config)
message = loom.schema_pb2.Config()
protobuf_dump(config, message)
with open_compressed(filename, 'wb') as f:
f.write(message.SerializeToString())
def test_predict(root, rows_csv, encoding, **unused):
COUNT = 10
with tempdir(cleanup_on_error=CLEANUP_ON_ERROR):
with loom.query.get_server(root, debug=True) as query_server:
result_out = 'predictions_out.csv'
rows_in = os.listdir(rows_csv)[0]
rows_in = os.path.join(rows_csv, rows_in)
encoders = json_load(encoding)
name_to_encoder = {e['name']: load_encoder(e) for e in encoders}
preql = loom.preql.PreQL(query_server, encoding)
preql.predict(rows_in, COUNT, result_out, id_offset=False)
with open_compressed(rows_in, 'rb') as fin:
with open(result_out, 'r') as fout:
in_reader = csv.reader(fin)
out_reader = csv.reader(fout)
fnames = in_reader.next()
out_reader.next()
for in_row in in_reader:
for i in range(COUNT):
out_row = out_reader.next()
bundle = zip(fnames, in_row, out_row)
for name, in_val, out_val in bundle:
encode = name_to_encoder[name]
observed = bool(in_val.strip())
if observed:
assert_almost_equal(
encode(in_val),
encode(out_val))
else:
assert_true(bool(out_val.strip()))
def _check_predictions(rows_in, result_out, encoding):
encoders = json_load(encoding)
name_to_encoder = {e['name']: load_encoder(e) for e in encoders}
with open_compressed(rows_in, 'rb') as fin:
with open(result_out, 'r') as fout:
in_reader = csv.reader(fin)
out_reader = csv.reader(fout)
fnames = in_reader.next()
out_reader.next()
for in_row in in_reader:
for i in range(COUNT):
out_row = out_reader.next()
bundle = zip(fnames, in_row, out_row)
for name, in_val, out_val in bundle:
if name == '_id':
assert_equal(in_val, out_val)
continue
encode = name_to_encoder[name]
observed = bool(in_val.strip())
if observed:
assert_almost_equal(
encode(in_val),
encode(out_val))
else:
assert_true(bool(out_val.strip()))
def config_dump(config, filename):
config = deepcopy(config)
fill_in_defaults(config)
message = loom.schema_pb2.Config()
protobuf_dump(config, message)
with open_compressed(filename, 'wb') as f:
f.write(message.SerializeToString())
def get_example_requests(model):
cross_cat = CrossCat()
with open_compressed(model) as f:
cross_cat.ParseFromString(f.read())
feature_count = sum(len(kind.featureids) for kind in cross_cat.kinds)
all_observed = [True] * feature_count
none_observed = [False] * feature_count
observeds = []
observeds.append(all_observed)
for f in xrange(feature_count):
observed = all_observed[:]
observed[f] = False
observeds.append(observed)
for f in xrange(feature_count):
observed = [sample_bernoulli(0.5) for _ in xrange(feature_count)]
observeds.append(observed)
for f in xrange(feature_count):
observed = none_observed[:]
observed[f] = True
observeds.append(observed)
observeds.append(none_observed)
requests = []
for i, observed in enumerate(observeds):
request = Query.Request()
request.id = "example-{}".format(i)
request.sample.data.observed[:] = none_observed
request.sample.to_sample[:] = observed
request.sample.sample_count = 1
requests.append(request)
return requests
def _import_rows_file(args):
encoding_in, rows_csv_in, rows_out, id_offset, id_stride = args
assert os.path.isfile(rows_csv_in)
encoders = json_load(encoding_in)
message = loom.cFormat.Row()
add_field = {
'booleans': message.add_booleans,
'counts': message.add_counts,
'reals': message.add_reals,
}
with open_compressed(rows_csv_in, 'rb') as f:
reader = csv.reader(f)
feature_names = list(reader.next())
name_to_pos = {name: i for i, name in enumerate(feature_names)}
schema = []
for encoder in encoders:
pos = name_to_pos.get(encoder['name'])
add = add_field[loom.schema.MODEL_TO_DATATYPE[encoder['model']]]
encode = load_encoder(encoder)
schema.append((pos, add, encode))
def rows():
for i, row in enumerate(reader):
message.id = id_offset + id_stride * i
for pos, add, encode in schema:
value = None if pos is None else row[pos].strip()
observed = bool(value)
message.add_observed(observed)
if observed:
add(encode(value))
yield message
message.Clear()
loom.cFormat.row_stream_dump(rows(), rows_out)
def make_fake_encoding(schema_in, model_in, encoding_out):
'''
Make a fake encoding from json schema + model.
Assume that feature names in schema correspond to featureids in model
e.g. schema was generated from loom.format.make_schema
'''
schema = json_load(schema_in)
fields = []
builders = []
name_to_builder = {}
for name, model in sorted(schema.iteritems()):
fields.append(loom.schema.MODEL_TO_DATATYPE[model])
Builder = FAKE_ENCODER_BUILDERS[model]
builder = Builder(name, model)
builders.append(builder)
name_to_builder[name] = builder
cross_cat = loom.schema_pb2.CrossCat()
with open_compressed(model_in, 'rb') as f:
cross_cat.ParseFromString(f.read())
for kind in cross_cat.kinds:
featureid = iter(kind.featureids)
for model in loom.schema.MODELS.iterkeys():
for shared in getattr(kind.product_model, model):
feature_name = '{:06d}'.format(featureid.next())
assert feature_name in schema
if model == 'dd':
for i in range(len(shared.alphas)):
name_to_builder[feature_name].add_value(str(i))
elif model == 'dpd':
for val in shared.values:
name_to_builder[feature_name].add_value(str(val))
encoders = [b.build() for b in builders]
ensure_fake_encoders_are_sorted(encoders)
json_dump(encoders, encoding_out)
def make_fake_encoding(schema_in, model_in, encoding_out):
'''
Make a fake encoding from json schema + model.
Assume that feature names in schema correspond to featureids in model
e.g. schema was generated from loom.format.make_schema
'''
schema = json_load(schema_in)
fields = []
builders = []
name_to_builder = {}
for name, model in sorted(schema.iteritems()):
fields.append(loom.schema.MODEL_TO_DATATYPE[model])
Builder = FAKE_ENCODER_BUILDERS[model]
builder = Builder(name, model)
builders.append(builder)
name_to_builder[name] = builder
cross_cat = loom.schema_pb2.CrossCat()
with open_compressed(model_in, 'rb') as f:
cross_cat.ParseFromString(f.read())
for kind in cross_cat.kinds:
featureid = iter(kind.featureids)
for model in loom.schema.MODELS.iterkeys():
for shared in getattr(kind.product_model, model):
feature_name = '{:06d}'.format(featureid.next())
assert feature_name in schema
if model == 'dd':
for i in range(len(shared.alphas)):
name_to_builder[feature_name].add_value(str(i))
elif model == 'dpd':
for val in shared.values:
name_to_builder[feature_name].add_value(str(val))
encoders = [b.build() for b in builders]
ensure_fake_encoders_are_sorted(encoders)
json_dump(encoders, encoding_out)
def ingest(name, schema=None, rows_csv=None, id_field=None, debug=False):
'''
Ingest dataset with optional json config.
Arguments:
name A unique identifier for ingest + inference
schema Json schema file, e.g., {"feature1": "nich"}
rows_csv File or directory of csv files or csv.gz files
id_field Column name of id field in input csv
debug Whether to run debug versions of C++ code
Environment variables:
LOOM_THREADS Number of concurrent ingest tasks
LOOM_VERBOSITY Verbosity level
'''
paths = loom.store.get_paths(name)
if schema is None:
schema = paths['ingest']['schema']
if rows_csv is None:
rows_csv = paths['ingest']['rows_csv']
if not os.path.exists(schema):
raise LoomError('Missing schema file: {}'.format(schema))
if not os.path.exists(rows_csv):
raise LoomError('Missing rows_csv file: {}'.format(rows_csv))
with open_compressed(paths['ingest']['version'], 'w') as f:
f.write(loom.__version__)
LOG('making schema row')
loom.format.make_schema_row(schema_in=schema,
schema_row_out=paths['ingest']['schema_row'])
LOG('making encoding')
loom.format.make_encoding(schema_in=schema,
rows_in=rows_csv,
encoding_out=paths['ingest']['encoding'])
LOG('importing rows')
loom.format.import_rows(encoding_in=paths['ingest']['encoding'],
rows_csv_in=rows_csv,
rows_out=paths['ingest']['rows'])
LOG('importing rowids')
loom.format.import_rowids(rows_csv_in=rows_csv,
rowids_out=paths['ingest']['rowids'],
id_field=id_field)
LOG('making tare rows')
loom.runner.tare(schema_row_in=paths['ingest']['schema_row'],
rows_in=paths['ingest']['rows'],
tares_out=paths['ingest']['tares'],
debug=debug)
tare_count = sum(1 for _ in protobuf_stream_load(paths['ingest']['tares']))
LOG('sparsifying rows WRT {} tare rows'.format(tare_count))
loom.runner.sparsify(schema_row_in=paths['ingest']['schema_row'],
tares_in=paths['ingest']['tares'],
rows_in=paths['ingest']['rows'],
rows_out=paths['ingest']['diffs'],
debug=debug)
loom.config.config_dump({}, paths['query']['config'])
def _loom_cross_cat(path, sample):
"""Return the loom CrossCat structure at `path`, whose id is `sample`."""
model_in = os.path.join(path, 'samples', 'sample.%d' % (sample, ),
'model.pb.gz')
cross_cat = loom.schema_pb2.CrossCat()
with open_compressed(model_in, 'rb') as f:
cross_cat.ParseFromString(f.read())
return cross_cat
def predict(self, rows_csv, count, result_out, id_offset=True):
with open_compressed(rows_csv, 'rb') as fin:
with open_compressed(result_out, 'w') as fout:
reader = csv.reader(fin)
writer = csv.writer(fout)
feature_names = list(reader.next())
writer.writerow(feature_names)
name_to_pos = {name: i for i, name in enumerate(feature_names)}
pos_to_decode = {}
schema = []
for encoder in self.encoders:
pos = name_to_pos.get(encoder['name'])
encode = load_encoder(encoder)
decode = load_decoder(encoder)
if pos is not None:
pos_to_decode[pos] = decode
schema.append((pos, encode))
for row in reader:
conditioning_row = []
to_sample = []
if id_offset:
row_id = row.pop(0)
for pos, encode, in schema:
value = None if pos is None else row[pos].strip()
observed = bool(value)
to_sample.append((not observed))
if observed is False:
conditioning_row.append(None)
else:
conditioning_row.append(encode(value))
samples = self.query_server.sample(
to_sample,
conditioning_row,
count)
for sample in samples:
if id_offset:
out_row = [row_id]
else:
out_row = []
for name in feature_names:
pos = name_to_pos[name]
decode = pos_to_decode[pos]
val = sample[pos]
out_row.append(val)
writer.writerow(out_row)
def csv_output(arg):
if arg is None:
outfile = StringIO()
yield CsvWriter(outfile, returns=outfile.getvalue)
elif hasattr(arg, 'write'):
yield CsvWriter(arg)
else:
with open_compressed(arg, 'w') as outfile:
yield CsvWriter(outfile)
def _get_cross_cat(self, bdb, generator_id, modelno):
"""Return the loom CrossCat structure whose id is `modelno`."""
model_in = os.path.join(
self._get_loom_project_path(bdb, generator_id),
'samples', 'sample.%d' % (modelno,), 'model.pb.gz')
cross_cat = loom.schema_pb2.CrossCat()
with open_compressed(model_in, 'rb') as f:
cross_cat.ParseFromString(f.read())
return cross_cat
def _get_cross_cat(self, bdb, generator_id, modelno):
"""Return the loom CrossCat structure whose id is `modelno`."""
model_in = os.path.join(self._get_loom_project_path(bdb, generator_id),
'samples', 'sample.%d' % (modelno, ),
'model.pb.gz')
cross_cat = loom.schema_pb2.CrossCat()
with open_compressed(model_in, 'rb') as f:
cross_cat.ParseFromString(f.read())
return cross_cat
def csv_output(arg):
if arg is None:
outfile = StringIO()
yield CsvWriter(outfile, returns=outfile.getvalue)
elif hasattr(arg, 'write'):
yield CsvWriter(arg)
else:
with open_compressed(arg, 'w') as outfile:
yield CsvWriter(outfile)
def group_sample((sample, featureid)):
model = CrossCat()
with open_compressed(sample['model']) as f:
model.ParseFromString(f.read())
for kindid, kind in enumerate(model.kinds):
if featureid in kind.featureids:
break
assignments = assignment_stream_load(sample['assign'])
return collate((a.groupids(kindid), a.rowid) for a in assignments)
def group_sample((sample, featureid)):
model = CrossCat()
with open_compressed(sample['model']) as f:
model.ParseFromString(f.read())
for kindid, kind in enumerate(model.kinds):
if featureid in kind.featureids:
break
assignments = assignment_stream_load(sample['assign'])
return collate((a.groupids(kindid), a.rowid) for a in assignments)
def generate_init(encoding_in, model_out, seed=0):
'''
Generate an initial model for inference.
'''
numpy.random.seed(seed)
encoders = json_load(encoding_in)
features = import_features(encoders)
cross_cat = generate_model(features)
with open_compressed(model_out, 'wb') as f:
f.write(cross_cat.SerializeToString())
def generate_init(encoding_in, model_out, seed=0):
'''
Generate an initial model for inference.
'''
numpy.random.seed(seed)
encoders = json_load(encoding_in)
features = import_features(encoders)
cross_cat = generate_model(features)
with open_compressed(model_out, 'wb') as f:
f.write(cross_cat.SerializeToString())
def ingest(name, schema='schema.json', rows_csv='rows.csv.gz', debug=False):
'''
Ingest dataset with optional json config.
Arguments:
name A unique identifier for ingest + inference
schema Json schema file, e.g., {"feature1": "nich"}
rows_csv File or directory of csv files
debug Whether to run debug versions of C++ code
Environment variables:
LOOM_THREADS Number of concurrent ingest tasks
LOOM_VERBOSITY Verbosity level
'''
if not os.path.exists(schema):
raise IOError('Missing schema file: {}'.format(schema))
if not os.path.exists(rows_csv):
raise IOError('Missing rows_csv file: {}'.format(rows_csv))
paths = loom.store.get_paths(name)
with open_compressed(paths['ingest']['version'], 'w') as f:
f.write(loom.__version__)
LOG('making schema row')
loom.format.make_schema_row(
schema_in=schema,
schema_row_out=paths['ingest']['schema_row'])
LOG('making encoding')
loom.format.make_encoding(
schema_in=schema,
rows_in=rows_csv,
encoding_out=paths['ingest']['encoding'])
LOG('importing rows')
loom.format.import_rows(
encoding_in=paths['ingest']['encoding'],
rows_csv_in=rows_csv,
rows_out=paths['ingest']['rows'])
LOG('making tare rows')
loom.runner.tare(
schema_row_in=paths['ingest']['schema_row'],
rows_in=paths['ingest']['rows'],
tares_out=paths['ingest']['tares'],
debug=debug)
tare_count = sum(1 for _ in protobuf_stream_load(paths['ingest']['tares']))
LOG('sparsifying rows WRT {} tare rows'.format(tare_count))
loom.runner.sparsify(
schema_row_in=paths['ingest']['schema_row'],
tares_in=paths['ingest']['tares'],
rows_in=paths['ingest']['rows'],
rows_out=paths['ingest']['diffs'],
debug=debug)
def make_schema_row(schema_in, schema_row_out):
'''
Convert json schema to protobuf schema row.
'''
schema = json_load(schema_in)
value = loom.schema_pb2.ProductValue()
value.observed.sparsity = loom.schema_pb2.ProductValue.Observed.DENSE
for model in schema.itervalues():
field = loom.schema.MODEL_TO_DATATYPE[model]
value.observed.dense.append(True)
getattr(value, field).append(EXAMPLE_VALUES[field])
with open_compressed(schema_row_out, 'wb') as f:
f.write(value.SerializeToString())
def test_infer(name, tares, shuffled, init, **unused):
with tempdir(cleanup_on_error=CLEANUP_ON_ERROR):
row_count = sum(1 for _ in protobuf_stream_load(shuffled))
with open_compressed(init) as f:
message = CrossCat()
message.ParseFromString(f.read())
kind_count = len(message.kinds)
for config in CONFIGS:
loom.config.fill_in_defaults(config)
schedule = config['schedule']
print 'config: {}'.format(config)
greedy = (schedule['extra_passes'] == 0)
kind_iters = config['kernels']['kind']['iterations']
kind_structure_is_fixed = greedy or kind_iters == 0
with tempdir(cleanup_on_error=CLEANUP_ON_ERROR):
config_in = os.path.abspath('config.pb.gz')
model_out = os.path.abspath('model.pb.gz')
groups_out = os.path.abspath('groups')
assign_out = os.path.abspath('assign.pbs.gz')
log_out = os.path.abspath('log.pbs.gz')
os.mkdir(groups_out)
loom.config.config_dump(config, config_in)
loom.runner.infer(
config_in=config_in,
rows_in=shuffled,
tares_in=tares,
model_in=init,
model_out=model_out,
groups_out=groups_out,
assign_out=assign_out,
log_out=log_out,
debug=True,)
if kind_structure_is_fixed:
assert_equal(len(os.listdir(groups_out)), kind_count)
group_counts = get_group_counts(groups_out)
assign_count = sum(1 for _ in protobuf_stream_load(assign_out))
assert_equal(assign_count, row_count)
print 'row_count: {}'.format(row_count)
print 'group_counts: {}'.format(' '.join(map(str, group_counts)))
for group_count in group_counts:
assert_true(
group_count <= row_count,
'groups are all singletons')
def test_infer(name, tares, shuffled, init, **unused):
with tempdir(cleanup_on_error=CLEANUP_ON_ERROR):
row_count = sum(1 for _ in protobuf_stream_load(shuffled))
with open_compressed(init) as f:
message = CrossCat()
message.ParseFromString(f.read())
kind_count = len(message.kinds)
for config in CONFIGS:
loom.config.fill_in_defaults(config)
schedule = config['schedule']
print 'config: {}'.format(config)
greedy = (schedule['extra_passes'] == 0)
kind_iters = config['kernels']['kind']['iterations']
kind_structure_is_fixed = greedy or kind_iters == 0
with tempdir(cleanup_on_error=CLEANUP_ON_ERROR):
config_in = os.path.abspath('config.pb.gz')
model_out = os.path.abspath('model.pb.gz')
groups_out = os.path.abspath('groups')
assign_out = os.path.abspath('assign.pbs.gz')
log_out = os.path.abspath('log.pbs.gz')
os.mkdir(groups_out)
loom.config.config_dump(config, config_in)
loom.runner.infer(
config_in=config_in,
rows_in=shuffled,
tares_in=tares,
model_in=init,
model_out=model_out,
groups_out=groups_out,
assign_out=assign_out,
log_out=log_out,
debug=True,
)
if kind_structure_is_fixed:
assert_equal(len(os.listdir(groups_out)), kind_count)
group_counts = get_group_counts(groups_out)
assign_count = sum(1 for _ in protobuf_stream_load(assign_out))
assert_equal(assign_count, row_count)
print 'row_count: {}'.format(row_count)
print 'group_counts: {}'.format(' '.join(map(str, group_counts)))
for group_count in group_counts:
assert_true(group_count <= row_count,
'groups are all singletons')
def create_dataset(row_count=ROW_COUNT):
'''
Extract dataset from image.
'''
scipy.misc.imsave(os.path.join(RESULTS, 'original.png'), IMAGE)
print 'sampling {} points from image'.format(row_count)
with open_compressed(SAMPLES, 'w') as f:
writer = csv.writer(f)
writer.writerow(['x', 'y'])
for row in sample_from_image(IMAGE, row_count):
writer.writerow(row)
with csv_reader(SAMPLES) as reader:
reader.next()
image = visualize_dataset(map(float, row) for row in reader)
scipy.misc.imsave(os.path.join(RESULTS, 'samples.png'), image)
def create_dataset(row_count=ROW_COUNT):
'''
Extract dataset from image.
'''
scipy.misc.imsave(os.path.join(RESULTS, 'original.png'), IMAGE)
print 'sampling {} points from image'.format(row_count)
with open_compressed(SAMPLES, 'w') as f:
writer = csv.writer(f)
writer.writerow(['x', 'y'])
for row in sample_from_image(IMAGE, row_count):
writer.writerow(row)
with csv_reader(SAMPLES) as reader:
reader.next()
image = visualize_dataset(map(float, row) for row in reader)
scipy.misc.imsave(os.path.join(RESULTS, 'samples.png'), image)
def import_rows():
row = Row()
pos = row.diff.pos
neg = row.diff.neg
pos.observed.sparsity = ProductValue.Observed.SPARSE
neg.observed.sparsity = ProductValue.Observed.SPARSE
with open_compressed(RAW) as infile:
doc_count = int(infile.next())
word_count = int(infile.next())
observed_count = int(infile.next())
print 'Importing {} observations of {} words in {} documents'.format(
observed_count,
word_count,
doc_count)
with open_compressed(DIFFS, 'wb') as outfile:
current_doc = None
for line in infile:
doc, feature, count = line.split()
if doc != current_doc:
if current_doc is not None:
pos.observed.sparse.sort()
neg.observed.sparse.sort()
protobuf_stream_write(row.SerializeToString(), outfile)
print_dot(every=1000)
current_doc = doc
row.id = int(doc)
del pos.booleans[:]
del pos.observed.sparse[:]
del neg.booleans[:]
del neg.observed.sparse[:]
feature = int(feature) - 1
pos.observed.sparse.append(feature)
pos.booleans.append(True)
neg.observed.sparse.append(feature)
neg.booleans.append(False)
protobuf_stream_write(row.SerializeToString(), outfile)
def generate(
feature_type='mixed',
row_count=1000,
feature_count=100,
density=0.5,
rows_out='rows.pbs.gz',
model_out='model.pb.gz',
groups_out=None,
assign_out=None,
init_out=None,
debug=False,
profile=None):
'''
Generate a synthetic dataset.
'''
root = os.getcwd()
rows_out = os.path.abspath(rows_out)
model_out = os.path.abspath(model_out)
if groups_out is not None:
groups_out = os.path.abspath(groups_out)
if assign_out is not None:
assign_out = os.path.abspath(assign_out)
if init_out is not None:
init_out = os.path.abspath(init_out)
features = generate_features(feature_count, feature_type)
model = generate_model(features)
with tempdir(cleanup_on_error=(not debug)):
if init_out is None:
init_out = os.path.abspath('init.pb.gz')
with open_compressed(init_out, 'wb') as f:
f.write(model.SerializeToString())
config = {'generate': {'row_count': row_count, 'density': density}}
config_in = os.path.abspath('config.pb.gz')
loom.config.config_dump(config, config_in)
os.chdir(root)
loom.runner.generate(
config_in=config_in,
model_in=init_out,
rows_out=rows_out,
model_out=model_out,
groups_out=groups_out,
assign_out=assign_out,
debug=debug,
profile=profile)
def make_schema(model_in, schema_out):
'''
Make a schema from a protobuf model.
'''
cross_cat = loom.schema_pb2.CrossCat()
with open_compressed(model_in, 'rb') as f:
cross_cat.ParseFromString(f.read())
schema = {}
for kind in cross_cat.kinds:
featureid = iter(kind.featureids)
for model in loom.schema.MODELS.iterkeys():
for shared in getattr(kind.product_model, model):
feature_name = '{:06d}'.format(featureid.next())
schema[feature_name] = model
json_dump(schema, schema_out)
return schema
def make_schema(model_in, schema_out):
'''
Make a schema from a protobuf model.
'''
cross_cat = loom.schema_pb2.CrossCat()
with open_compressed(model_in, 'rb') as f:
cross_cat.ParseFromString(f.read())
schema = {}
for kind in cross_cat.kinds:
featureid = iter(kind.featureids)
for model in loom.schema.MODELS.iterkeys():
for shared in getattr(kind.product_model, model):
feature_name = '{:06d}'.format(featureid.next())
schema[feature_name] = model
json_dump(schema, schema_out)
return schema
def generate(feature_type='mixed',
row_count=1000,
feature_count=100,
density=0.5,
rows_out='rows.pbs.gz',
model_out='model.pb.gz',
groups_out=None,
assign_out=None,
init_out=None,
debug=False,
profile=None):
'''
Generate a synthetic dataset.
'''
root = os.getcwd()
rows_out = os.path.abspath(rows_out)
model_out = os.path.abspath(model_out)
if groups_out is not None:
groups_out = os.path.abspath(groups_out)
if assign_out is not None:
assign_out = os.path.abspath(assign_out)
if init_out is not None:
init_out = os.path.abspath(init_out)
features = generate_features(feature_count, feature_type)
model = generate_model(features)
with tempdir(cleanup_on_error=(not debug)):
if init_out is None:
init_out = os.path.abspath('init.pb.gz')
with open_compressed(init_out, 'wb') as f:
f.write(model.SerializeToString())
config = {'generate': {'row_count': row_count, 'density': density}}
config_in = os.path.abspath('config.pb.gz')
loom.config.config_dump(config, config_in)
os.chdir(root)
loom.runner.generate(config_in=config_in,
model_in=init_out,
rows_out=rows_out,
model_out=model_out,
groups_out=groups_out,
assign_out=assign_out,
debug=debug,
profile=profile)
def make_schema_row(schema_in, schema_row_out):
'''
Convert json schema to protobuf schema row.
'''
schema = json_load(schema_in)
if not schema:
raise LoomError('Schema is empty: {}'.format(schema_in))
value = loom.schema_pb2.ProductValue()
value.observed.sparsity = loom.schema_pb2.ProductValue.Observed.DENSE
for model in schema.itervalues():
try:
field = loom.schema.MODEL_TO_DATATYPE[model]
except KeyError:
raise LoomError('Unknown model {} in schema {}'.format(
model, schema_in))
value.observed.dense.append(True)
getattr(value, field).append(EXAMPLE_VALUES[field])
with open_compressed(schema_row_out, 'wb') as f:
f.write(value.SerializeToString())
def make_schema_row(schema_in, schema_row_out):
'''
Convert json schema to protobuf schema row.
'''
schema = json_load(schema_in)
if not schema:
raise LoomError('Schema is empty: {}'.format(schema_in))
value = loom.schema_pb2.ProductValue()
value.observed.sparsity = loom.schema_pb2.ProductValue.Observed.DENSE
for model in schema.itervalues():
try:
field = loom.schema.MODEL_TO_DATATYPE[model]
except KeyError:
raise LoomError('Unknown model {} in schema {}'.format(
model, schema_in))
value.observed.dense.append(True)
getattr(value, field).append(EXAMPLE_VALUES[field])
with open_compressed(schema_row_out, 'wb') as f:
f.write(value.SerializeToString())
def test_predict_pandas(root, rows_csv, schema, **unused):
feature_count = len(json_load(schema))
with loom.preql.get_server(root, debug=True) as preql:
rows_filename = os.path.join(rows_csv, os.listdir(rows_csv)[0])
with open_compressed(rows_filename) as f:
rows_df = pandas.read_csv(f,
converters=preql.converters,
index_col='_id')
print 'rows_df ='
print rows_df
row_count = rows_df.shape[0]
assert_equal(rows_df.shape[1], feature_count)
rows_io = StringIO(rows_df.to_csv())
result_string = preql.predict(rows_io, COUNT, id_offset=True)
result_df = pandas.read_csv(StringIO(result_string), index_col=False)
print 'result_df ='
print result_df
assert_equal(result_df.ndim, 2)
assert_equal(result_df.shape[0], row_count * COUNT)
assert_equal(result_df.shape[1], 1 + feature_count)
def test_predict_pandas(root, rows_csv, schema, **unused):
feature_count = len(json_load(schema))
with loom.preql.get_server(root, debug=True) as preql:
rows_filename = os.path.join(rows_csv, os.listdir(rows_csv)[0])
with open_compressed(rows_filename) as f:
rows_df = pandas.read_csv(
f,
converters=preql.converters,
index_col='_id')
print 'rows_df ='
print rows_df
row_count = rows_df.shape[0]
assert_equal(rows_df.shape[1], feature_count)
rows_io = StringIO(rows_df.to_csv())
result_string = preql.predict(rows_io, COUNT, id_offset=True)
result_df = pandas.read_csv(StringIO(result_string), index_col=False)
print 'result_df ='
print result_df
assert_equal(result_df.ndim, 2)
assert_equal(result_df.shape[0], row_count * COUNT)
assert_equal(result_df.shape[1], 1 + feature_count)
def _make_encoder_builders_file((schema_in, rows_in)):
assert os.path.isfile(rows_in)
schema = json_load(schema_in)
with open_compressed(rows_in, 'rb') as f:
reader = csv.reader(f)
header = reader.next()
builders = []
for name in header:
if name in schema:
model = schema[name]
Builder = ENCODER_BUILDERS[model]
builder = Builder(name, model)
else:
builder = None
builders.append(builder)
for row in reader:
for value, builder in izip(row, builders):
if builder is not None:
value = value.strip()
if value:
builder.add_value(value)
return [b for b in builders if b is not None]
def generate(
feature_type='mixed',
row_count=1000,
feature_count=100,
density=0.5,
init_out='init.pb.gz',
rows_out='rows.pbs.gz',
model_out='model.pb.gz',
groups_out='groups',
debug=False,
profile=None):
'''
Generate a synthetic dataset.
'''
root = os.path.abspath(os.path.curdir)
init_out = os.path.abspath(init_out)
rows_out = os.path.abspath(rows_out)
model_out = os.path.abspath(model_out)
groups_out = os.path.abspath(groups_out)
model = generate_model(row_count, feature_count, feature_type, density)
with open_compressed(init_out, 'w') as f:
f.write(model.SerializeToString())
with tempdir(cleanup_on_error=(not debug)):
config = {'generate': {'row_count': row_count, 'density': density}}
config_in = os.path.abspath('config.pb.gz')
loom.config.config_dump(config, config_in)
os.chdir(root)
loom.runner.generate(
config_in=config_in,
model_in=init_out,
rows_out=rows_out,
model_out=model_out,
groups_out=groups_out,
debug=debug,
profile=profile)
def export_rows(encoding_in, rows_in, rows_csv_out, chunk_size=1000000):
'''
Export rows from protobuf stream to csv.
'''
for ext in ['.csv', '.gz', '.bz2']:
assert not rows_csv_out.endswith(ext),\
'rows_csv_out should be a dirname'
assert chunk_size > 0
encoders = json_load(encoding_in)
fields = [loom.schema.MODEL_TO_DATATYPE[e['model']] for e in encoders]
decoders = [load_decoder(e) for e in encoders]
header = [e['name'] for e in encoders]
if os.path.exists(rows_csv_out):
shutil.rmtree(rows_csv_out)
os.makedirs(rows_csv_out)
rows = loom.cFormat.row_stream_load(rows_in)
try:
empty = None
for i in xrange(MAX_CHUNK_COUNT):
file_out = os.path.join(
rows_csv_out,
'rows_{:06d}.csv.gz'.format(i))
with open_compressed(file_out, 'wb') as f:
writer = csv.writer(f)
writer.writerow(header)
empty = file_out
for j in xrange(chunk_size):
data = rows.next().iter_data()
schema = izip(data['observed'], fields, decoders)
row = [
decode(data[field].next()) if observed else ''
for observed, field, decode in schema
]
writer.writerow(row)
empty = None
except StopIteration:
if empty:
os.remove(empty)
def relate(self, columns, result_out, sample_count=1000):
"""
Compute pairwise related scores between all pairs of
columns in columns.
Related scores are defined to be:
Related(X, Y) = I(X; Y) / H(X, Y)
Where:
I(X; Y) is the mutual information between X and Y:
I(X; Y) = E[ log( p(x, y)) / ( p(x) p(y) ) ]; x, y ~ p(x, y)
H(X) is the entropy of X:
H(X) = E[ log( p(x) )]; x ~ p(x)
Expectations are estimated via monte carlo with `sample_count` samples
"""
with open_compressed(result_out, 'w') as f:
writer = csv.writer(f)
writer.writerow(self.feature_names)
for target_column in set(columns):
out_row = [target_column]
to_sample1 = self.cols_to_sample([target_column])
for to_relate in self.feature_names:
to_sample2 = self.cols_to_sample([to_relate])
mi = self.query_server.mutual_information(
to_sample1,
to_sample2,
sample_count=sample_count).mean
joined = [to_relate, target_column]
to_sample_both = self.cols_to_sample(joined)
joint_entropy = self.query_server.entropy(
to_sample_both,
sample_count=sample_count).mean
normalized_mi = self.normalize_mutual_information(
mi,
joint_entropy)
out_row.append(normalized_mi)
writer.writerow(out_row)
def import_schema():
schema = [line.strip() for line in open_compressed(VOCAB)]
with open_compressed(SCHEMA_CSV, 'w') as outfile:
writer = csv.writer(outfile)
writer.writerow(schema)
return schema
def csv_writer(filename):
with open_compressed(filename, 'wb') as f:
yield csv.writer(f)
def pickle_dump(data, filename):
with open_compressed(filename, 'wb') as f:
pickle.dump(data, f)
def pickle_load(filename):
with open_compressed(filename, 'rb') as f:
return pickle.load(f)
def _load_checkpoint(step):
message = loom.schema_pb2.Checkpoint()
filename = checkpoint_files(step)['checkpoint']
with open_compressed(filename) as f:
message.ParseFromString(f.read())
return message
def csv_input(arg):
if hasattr(arg, 'read'):
yield csv.reader(arg)
else:
with open_compressed(arg, 'rb') as infile:
yield csv.reader(infile)
def get_example_requests(model, rows, query_type='mixed'):
assert query_type in ['sample', 'score', 'mixed']
cross_cat = CrossCat()
with open_compressed(model, 'rb') as f:
cross_cat.ParseFromString(f.read())
feature_count = sum(len(kind.featureids) for kind in cross_cat.kinds)
featureids = range(feature_count)
nontrivials = [True] * feature_count
for kind in cross_cat.kinds:
fs = iter(kind.featureids)
for model in loom.schema.MODELS.iterkeys():
for shared in getattr(kind.product_model, model):
f = fs.next()
if model == 'dd':
if len(shared.alphas) == 0:
nontrivials[f] = False
elif model == 'dpd':
if len(shared.betas) == 0:
nontrivials[f] = False
all_observed = nontrivials[:]
none_observed = [False] * feature_count
observeds = []
observeds.append(all_observed)
for f, nontrivial in izip(featureids, nontrivials):
if nontrivial:
observed = all_observed[:]
observed[f] = False
observeds.append(observed)
for f in featureids:
observed = [
nontrivial and sample_bernoulli(0.5)
for nontrivial in nontrivials
]
observeds.append(observed)
for f, nontrivial in izip(featureids, nontrivials):
if nontrivial:
observed = none_observed[:]
observed[f] = True
observeds.append(observed)
observeds.append(none_observed)
requests = []
for i, observed in enumerate(observeds):
request = Query.Request()
request.id = "example-{}".format(i)
if query_type in ['sample', 'mixed']:
set_diff(request.sample.data, none_observed)
request.sample.to_sample.sparsity = DENSE
request.sample.to_sample.dense[:] = observed
request.sample.sample_count = 1
if query_type in ['score', 'mixed']:
set_diff(request.score.data, none_observed)
requests.append(request)
for row in load_rows(rows)[:20]:
i += 1
request = Query.Request()
request.id = "example-{}".format(i)
if query_type in ['sample', 'mixed']:
request.sample.sample_count = 1
request.sample.data.MergeFrom(row.diff)
request.sample.to_sample.sparsity = DENSE
conditions = izip(nontrivials, row.diff.pos.observed.dense)
to_sample = [
nontrivial and not is_observed
for nontrivial, is_observed in conditions
]
set_observed(request.sample.to_sample, to_sample)
if query_type in ['score', 'mixed']:
request.score.data.MergeFrom(row.diff)
requests.append(request)
return requests
def import_schema():
schema = [line.strip() for line in open_compressed(VOCAB)]
with open_compressed(SCHEMA_CSV, 'w') as outfile:
writer = csv.writer(outfile)
writer.writerow(schema)
return schema
def csv_reader(filename):
with open_compressed(filename, 'rb') as f:
yield csv.reader(f)
def csv_load(filename):
with open_compressed(filename) as f:
reader = csv.reader(f)
return list(reader)
def generate_one((name, sample_count, force, debug)):
paths = loom.store.get_paths(name, sample_count=sample_count)
if not force and all(os.path.exists(f) for f in paths.itervalues()):
with open_compressed(paths['ingest']['version']) as f:
version = f.read().strip()
if version == loom.__version__:
return
print 'generating', name
mkdir_p(paths['root'])
with open_compressed(paths['ingest']['version'], 'w') as f:
f.write(loom.__version__)
config = CONFIGS[name]
chunk_size = max(10, (config['row_count'] + 7) / 8)
loom.transforms.make_fake_transforms(
transforms_out=paths['ingest']['transforms'])
loom.generate.generate(
init_out=paths['samples'][0]['init'],
rows_out=paths['ingest']['rows'],
model_out=paths['samples'][0]['model'],
groups_out=paths['samples'][0]['groups'],
assign_out=paths['samples'][0]['assign'],
**config)
loom.format.make_schema(
model_in=paths['samples'][0]['model'],
schema_out=paths['ingest']['schema'])
loom.format.make_fake_encoding(
schema_in=paths['ingest']['schema'],
model_in=paths['samples'][0]['model'],
encoding_out=paths['ingest']['encoding'])
loom.format.make_schema_row(
schema_in=paths['ingest']['schema'],
schema_row_out=paths['ingest']['schema_row'])
loom.runner.tare(
schema_row_in=paths['ingest']['schema_row'],
rows_in=paths['ingest']['rows'],
tares_out=paths['ingest']['tares'],
debug=debug)
loom.runner.sparsify(
schema_row_in=paths['ingest']['schema_row'],
tares_in=paths['ingest']['tares'],
rows_in=paths['ingest']['rows'],
rows_out=paths['ingest']['diffs'],
debug=debug)
loom.format.export_rows(
encoding_in=paths['ingest']['encoding'],
rows_in=paths['ingest']['rows'],
rows_csv_out=paths['ingest']['rows_csv'],
chunk_size=chunk_size)
loom.format.import_rowids(
rows_csv_in=paths['ingest']['rows_csv'],
rowids_out=paths['ingest']['rowids'],
id_field='_id')
protobuf_stream_dump([], paths['query']['query_log'])
loom.config.config_dump({}, paths['query']['config'])
for seed, sample in enumerate(paths['samples']):
loom.config.config_dump({'seed': seed}, sample['config'])
loom.generate.generate_init(
encoding_in=paths['ingest']['encoding'],
model_out=sample['init'],
seed=seed)
loom.runner.shuffle(
rows_in=paths['ingest']['diffs'],
rows_out=sample['shuffled'],
seed=seed,
debug=debug)
protobuf_stream_dump([], sample['infer_log'])
sample0 = paths['samples'][0]
for seed, sample in enumerate(paths['samples'][1:]):
if LOOM_DEBUG_MIX:
cp_ns(sample0['model'], sample['model'])
cp_ns(sample0['groups'], sample['groups'])
cp_ns(sample0['assign'], sample['assign'])
else:
loom.runner.mix(
config_in=sample['config'],
rows_in=paths['ingest']['rows'],
model_in=sample0['model'],
groups_in=sample0['groups'],
assign_in=sample0['assign'],
model_out=sample['model'],
groups_out=sample['groups'],
assign_out=sample['assign'],
debug=debug)
loom.consensus.make_fake_consensus(
paths=paths,
debug=debug)
def get_example_requests(model, rows, query_type='mixed'):
assert query_type in ['sample', 'score', 'mixed']
cross_cat = CrossCat()
with open_compressed(model, 'rb') as f:
cross_cat.ParseFromString(f.read())
feature_count = sum(len(kind.featureids) for kind in cross_cat.kinds)
featureids = range(feature_count)
nontrivials = [True] * feature_count
for kind in cross_cat.kinds:
fs = iter(kind.featureids)
for model in loom.schema.MODELS.iterkeys():
for shared in getattr(kind.product_model, model):
f = fs.next()
if model == 'dd':
if len(shared.alphas) == 0:
nontrivials[f] = False
elif model == 'dpd':
if len(shared.betas) == 0:
nontrivials[f] = False
all_observed = nontrivials[:]
none_observed = [False] * feature_count
observeds = []
observeds.append(all_observed)
for f, nontrivial in izip(featureids, nontrivials):
if nontrivial:
observed = all_observed[:]
observed[f] = False
observeds.append(observed)
for f in featureids:
observed = [
nontrivial and sample_bernoulli(0.5)
for nontrivial in nontrivials
]
observeds.append(observed)
for f, nontrivial in izip(featureids, nontrivials):
if nontrivial:
observed = none_observed[:]
observed[f] = True
observeds.append(observed)
observeds.append(none_observed)
requests = []
for i, observed in enumerate(observeds):
request = Query.Request()
request.id = "example-{}".format(i)
if query_type in ['sample', 'mixed']:
set_diff(request.sample.data, none_observed)
request.sample.to_sample.sparsity = DENSE
request.sample.to_sample.dense[:] = observed
request.sample.sample_count = 1
if query_type in ['score', 'mixed']:
set_diff(request.score.data, none_observed)
requests.append(request)
for row in load_rows(rows)[:20]:
i += 1
request = Query.Request()
request.id = "example-{}".format(i)
if query_type in ['sample', 'mixed']:
request.sample.sample_count = 1
request.sample.data.MergeFrom(row.diff)
request.sample.to_sample.sparsity = DENSE
conditions = izip(nontrivials, row.diff.pos.observed.dense)
to_sample = [
nontrivial and not is_observed
for nontrivial, is_observed in conditions
]
set_observed(request.sample.to_sample, to_sample)
if query_type in ['score', 'mixed']:
request.score.data.MergeFrom(row.diff)
requests.append(request)
return requests
def csv_dump(data, filename):
with open_compressed(filename, 'w') as f:
writer = csv.writer(f)
for row in data:
writer.writerow(row)
