vae_params = {
'n_code': 50,
# VAE results get worse with more epochs in preliminary optimization
#(Pumed with threshold 50)
'n_epochs': 50,
'batch_size': 500,
'n_hidden': 100,
'normalize_inputs': True,
}
# Models without metadata
BASELINES = [
# RandomBaseline(),
# MostPopular(),
Countbased(),
SVDRecommender(1000, use_title=False),
]
RECOMMENDERS = [
AAERecommender(adversarial=False, lr=0.001, **ae_params),
AAERecommender(prior='gauss', gen_lr=0.001, reg_lr=0.001, **ae_params),
VAERecommender(conditions=None, **vae_params),
DAERecommender(conditions=None, **ae_params)
]
# Metadata to use
CONDITIONS = ConditionList([
('title', PretrainedWordEmbeddingCondition(VECTORS)),
# ('author', CategoricalCondition(embedding_dim=32, reduce="sum",
# sparse=True, embedding_on_gpu=True))
])
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
'--model',
type=str,
defaule='aae',
# All possible method should appear here
choices=['cm', 'svd', 'ae', 'aae', 'mlp'],
help="Specify the model to use [aae]")
parser.add_argument('--epochs',
type=int,
default=20,
help="Specify the number of training epochs [50]")
parser.add_argument('--hidden',
type=int,
default=200,
help="Number of hidden units [100]")
parser.add_argument('--no-title',
action='store_false',
default=True,
dest='use_title',
help="Do not use the playlist titles")
parser.add_argument('--max-items',
type=int,
default=75000,
help="Limit the max number of considered items")
parser.add_argument(
'--vocab-size',
type=int,
default=50000,
help="Limit the max number of distinct condition words")
parser.add_argument('-j',
'--jobs',
type=int,
default=4,
help="Number of jobs for data loading [4].")
parser.add_argument('-o',
'--outfile',
default="submission.csv",
type=str,
help="Write submissions to this path")
parser.add_argument('--use-embedding',
default=False,
action='store_true',
help="Use embedding (SGNS GoogleNews) [false]")
parser.add_argument('--dont-aggregate',
action='store_false',
dest='aggregate',
default=True,
help="Aggregate track metadata as side info input")
parser.add_argument('--debug',
action='store_true',
default=False,
help="Activate debug mode, run only on small sample")
parser.add_argument(
'-x',
'--exclude',
type=argparse.FileType('r'),
default=None,
help="Path to file with slice filenames to exclude for training")
parser.add_argument(
'--dev',
type=str,
default=None,
help='Path to dev set, use in combination with (-x, --exclude)')
parser.add_argument('--no-idf',
action='store_false',
default=True,
dest='use_idf',
help="Do **not** use idf re-weighting")
parser.add_argument('--lr',
type=float,
default=0.001,
help="Initial learning rate [0.001]")
parser.add_argument('--code',
type=int,
default=100,
help="Code dimension [50]")
args = parser.parse_args()
# Either exclude and dev set, or no exclude and test set
assert (args.dev is None) == (args.exclude is None)
if args.dev is not None:
print("Making submission for dev set:", args.dev)
assert os.path.isfile(args.dev)
# Dump args into submission file
if os.path.exists(args.outfile) and \
input("Path '{}' exists. Overwrite? [y/N]"
.format(args.outfile)) != 'y':
exit(-1)
with open(args.outfile, 'w') as out:
print('#', args, file=out)
if args.use_embedding:
print("Loading embedding:", W2V_PATH)
vectors = KeyedVectors.load_word2vec_format(W2V_PATH,
binary=W2V_IS_BINARY)
else:
vectors = None
# Create the model as specified by command line args
# Count-based never uses title
# Decoding recommender always uses title
tfidf_params = {'max_features': args.vocab_size, 'use_idf': args.use_idf}
model = {
'cm':
Countbased(),
'svd':
SVDRecommender(use_title=args.use_title),
'ae':
AAERecommender(use_title=args.use_title,
adversarial=False,
n_hidden=args.hidden,
n_code=args.code,
n_epochs=args.epochs,
embedding=vectors,
lr=args.lr,
tfidf_params=tfidf_params),
'aae':
AAERecommender(
use_title=args.use_title,
adversarial=True,
n_hidden=args.hidden,
n_code=args.code,
n_epochs=args.epochs,
gen_lr=args.lr,
reg_lr=args.lr, # same gen and reg lrs
embedding=vectors,
tfidf_params=tfidf_params),
'mlp':
DecodingRecommender(n_epochs=args.epochs,
n_hidden=args.hidden,
embedding=vectors,
tfidf_params=tfidf_params)
}[args.model]
track_attrs = TRACK_INFO if args.aggregate else None
if args.exclude is not None:
# Dev set case, exclude dev set data
exclude = [line.strip() for line in args.exclude]
else:
# Real submission case, do not exclude any training data
exclude = None
# = Training =
print("Loading data from {} using {} jobs".format(DATA_PATH, args.jobs))
playlists = playlists_from_slices(DATA_PATH,
n_jobs=args.jobs,
debug=args.debug,
without=exclude)
print("Unpacking playlists")
train_set = Bags(*unpack_playlists(playlists, aggregate=track_attrs))
print("Building vocabulary of {} most frequent items".format(
args.max_items))
vocab, __counts = train_set.build_vocab(max_features=args.max_items,
apply=False)
train_set = train_set.apply_vocab(vocab)
print("Training set:", train_set, sep='\n')
print("Training for {} epochs".format(args.epochs))
try:
model.train(train_set)
except KeyboardInterrupt:
print("Training interrupted by keyboard, pass.")
# Not required anymore
del train_set
# = Predictions =
if args.dev is not None:
print("Loading and unpacking DEV set")
data, index2playlist, side_info = unpack_playlists(
load(args.dev), aggregate=track_attrs)
else:
print("Loading and unpacking test set")
data, index2playlist, side_info = unpack_playlists(
load(TEST_PATH), aggregate=track_attrs)
test_set = Bags(data, index2playlist, side_info)
# Apply same vocabulary as in training
test_set = test_set.apply_vocab(vocab)
print("Test set:", test_set, sep='\n')
pred = model.predict(test_set)
if sp.issparse(pred):
pred = pred.toarray()
else:
pred = np.asarray(pred)
print("Scaling and removing non-missing items")
pred = remove_non_missing(pred, test_set.tocsr(), copy=False)
index2trackid = {v: k for k, v in vocab.items()}
print("Making submission:", args.outfile)
make_submission(pred, index2playlist, index2trackid, outfile=args.outfile)
print("Success.")
print("Make sure to verify the submission format via", VERIFY_SCRIPT)
def main(year,
dataset,
min_count=None,
outfile=None,
drop=1,
baselines=False,
autoencoders=False,
conditioned_autoencoders=False,
all_metadata=True):
""" Main function for training and evaluating AAE methods on DBLP data """
assert baselines or autoencoders or conditioned_autoencoders, "Please specify what to run"
if all_metadata:
# V2 - all metadata
CONDITIONS = ConditionList([
('title', PretrainedWordEmbeddingCondition(VECTORS)),
('venue', PretrainedWordEmbeddingCondition(VECTORS)),
(
'author',
CategoricalCondition(
embedding_dim=32,
reduce="sum", # vocab_size=0.01,
sparse=False,
embedding_on_gpu=True))
])
else:
# V1 - only title metadata
CONDITIONS = ConditionList([
('title', PretrainedWordEmbeddingCondition(VECTORS))
])
#### CONDITOINS defined
ALL_MODELS = []
if baselines:
# Models without metadata
BASELINES = [
# RandomBaseline(),
# MostPopular(),
Countbased(),
SVDRecommender(1000, use_title=False)
]
ALL_MODELS += BASELINES
if not all_metadata:
# SVD can use only titles not generic conditions
ALL_MODELS += [SVDRecommender(1000, use_title=True)]
if autoencoders:
AUTOENCODERS = [
AAERecommender(adversarial=False,
conditions=None,
lr=0.001,
**AE_PARAMS),
AAERecommender(adversarial=True,
conditions=None,
gen_lr=0.001,
reg_lr=0.001,
**AE_PARAMS),
VAERecommender(conditions=None, **AE_PARAMS),
DAERecommender(conditions=None, **AE_PARAMS)
]
ALL_MODELS += AUTOENCODERS
if conditioned_autoencoders:
# Model with metadata (metadata used as set in CONDITIONS above)
CONDITIONED_AUTOENCODERS = [
AAERecommender(adversarial=False,
conditions=CONDITIONS,
lr=0.001,
**AE_PARAMS),
AAERecommender(adversarial=True,
conditions=CONDITIONS,
gen_lr=0.001,
reg_lr=0.001,
**AE_PARAMS),
DecodingRecommender(CONDITIONS,
n_epochs=100,
batch_size=1000,
optimizer='adam',
n_hidden=100,
lr=0.001,
verbose=True),
VAERecommender(conditions=CONDITIONS, **AE_PARAMS),
DAERecommender(conditions=CONDITIONS, **AE_PARAMS)
]
ALL_MODELS += CONDITIONED_AUTOENCODERS
print("Finished preparing models:", *ALL_MODELS, sep='\n\t')
path = DATA_PATH + ("dblp-ref/" if dataset == "dblp" else "acm.txt")
print("Loading data from", path)
papers = papers_from_files(path, dataset, n_jobs=4)
print("Unpacking {} data...".format(dataset))
bags_of_papers, ids, side_info = unpack_papers(papers)
del papers
bags = Bags(bags_of_papers, ids, side_info)
if args.compute_mi:
from aaerec.utils import compute_mutual_info
print("[MI] Dataset:", dataset)
print("[MI] min Count:", min_count)
tmp = bags.build_vocab(min_count=min_count, max_features=None)
mi = compute_mutual_info(tmp,
conditions=None,
include_labels=True,
normalize=True)
with open('mi.csv', 'a') as mifile:
print(dataset, min_count, mi, sep=',', file=mifile)
print("=" * 78)
exit(0)
log("Whole dataset:", logfile=outfile)
log(bags, logfile=outfile)
evaluation = Evaluation(bags, year, logfile=outfile)
evaluation.setup(min_count=min_count, min_elements=2, drop=drop)
with open(outfile, 'a') as fh:
print("~ Partial List + Titles + Author + Venue", "~" * 42, file=fh)
evaluation(ALL_MODELS, batch_size=1000)