def main(args):
dataset_path = Path(args.dataset) # list current absolute path
if not dataset_path.exists():
raise FileNotFoundError(f'Dataset not found: {args.dataset}')
dataset = data.Dataset(dataset_path)
dataset_name = dataset_path.name # last part of the path
if args.method == 'datewise':
resources = Path(args.resources)
models_path = resources / 'supervised_date_ranker.{}.pkl'.format(
dataset_name)
# load regression models for date ranking
key_to_model = utils.load_pkl(models_path)
date_ranker = datewise.SupervisedDateRanker(method='regression')
sent_collector = datewise.PM_Mean_SentenceCollector(clip_sents=5,
pub_end=2)
summarizer = summarizers.CentroidOpt()
system = datewise.DatewiseTimelineGenerator(
date_ranker=date_ranker,
summarizer=summarizer,
sent_collector=sent_collector,
key_to_model=key_to_model)
elif args.method == 'clust':
cluster_ranker = clust.ClusterDateMentionCountRanker()
#clusterer = clust.AffinityPropagationClusterer()
clusterer = clust.TemporalMarkovClusterer()
summarizer = summarizers.Pegasus()
system = clust.ClusteringTimelineGenerator(
cluster_ranker=cluster_ranker,
clusterer=clusterer,
summarizer=summarizer,
clip_sents=5,
unique_dates=True,
)
else:
raise ValueError(f'Method not found: {args.method}')
if dataset_name == 'entities':
evaluate(system,
dataset,
args.output,
trunc_timelines=True,
time_span_extension=7)
else:
evaluate(system,
dataset,
args.output,
trunc_timelines=False,
time_span_extension=0)
def main(args):
dataset_path = Path(args.dataset)
if not dataset_path.exists():
raise FileNotFoundError(f'Dataset not found: {args.dataset}')
dataset = data.Dataset(dataset_path)
dataset_name = dataset_path.name
if args.method == 'datewise':
# load regression models for date ranking
key_to_model = utils.load_pkl(args.model)
models = list(key_to_model.values())
date_ranker = datewise.SupervisedDateRanker(method='regression')
# there are multiple models (for cross-validation),
# we just an arbitrary model, the first one
date_ranker.model = models[0]
sent_collector = datewise.PM_Mean_SentenceCollector(
clip_sents=2, pub_end=2)
summarizer = summarizers.CentroidOpt()
system = datewise.DatewiseTimelineGenerator(
date_ranker=date_ranker,
summarizer=summarizer,
sent_collector=sent_collector,
key_to_model = key_to_model
)
elif args.method == 'clust':
cluster_ranker = clust.ClusterDateMentionCountRanker()
clusterer = clust.TemporalMarkovClusterer()
summarizer = summarizers.CentroidOpt()
system = clust.ClusteringTimelineGenerator(
cluster_ranker=cluster_ranker,
clusterer=clusterer,
summarizer=summarizer,
clip_sents=2,
unique_dates=True,
)
else:
raise ValueError(f'Method not found: {args.method}')
run(system, dataset, args.output)
def main(args):
dataset_path = Path(args.dataset)
if not dataset_path.exists():
raise FileNotFoundError(f'Dataset not found: {args.dataset}')
dataset = data.Dataset(dataset_path)
dataset_name = dataset_path.name
if args.method == 'datewise':
resources = Path(args.resources)
# load date_models for date ranking
if args.model == 'new_lr':
method = 'log_regression'
model_path = resources / 'date_ranker_new_lr.all.pkl'
key_to_model = utils.load_pkl(model_path)
model = key_to_model[dataset_name]
elif is_neural_net(args.model): # fcn, deep_fcn, cnn, wide_fcn
model, model_path = model_selector(args.model)
model.load_state_dict(torch.load(model_path)[dataset_name])
method = 'neural_net'
key_to_model = None
else:
method = 'linear_regression'
model_path = resources / 'date_ranker_orig.{}.pkl'.format(
dataset_name)
key_to_model = utils.load_pkl(model_path)
model = None
date_ranker = datewise.SupervisedDateRanker(model, method=method)
sent_collector = datewise.PM_Mean_SentenceCollector(clip_sents=5,
pub_end=2)
summarizer = summarizers.CentroidOpt()
system = datewise.DatewiseTimelineGenerator(
date_ranker=date_ranker,
summarizer=summarizer,
sent_collector=sent_collector,
key_to_model=key_to_model,
method=method)
elif args.method == 'clust':
cluster_ranker = clust.ClusterDateMentionCountRanker()
clusterer = clust.TemporalMarkovClusterer(max_days=1)
summarizer = summarizers.CentroidOpt()
system = clust.ClusteringTimelineGenerator(
cluster_ranker=cluster_ranker,
clusterer=clusterer,
summarizer=summarizer,
clip_sents=5,
unique_dates=True,
)
elif args.method == 'clust_sbertsum':
cluster_ranker = clust.ClusterDateMentionCountRanker()
clusterer = clust.TemporalMarkovClusterer()
summarizer = summarizers.SBERTSummarizer(date_only=False,
summary_criteria='similarity',
candidate_sents_per=5,
compare_with='both')
system = clust.ClusteringTimelineGenerator(
cluster_ranker=cluster_ranker,
clusterer=clusterer,
summarizer=summarizer,
clip_sents=5,
unique_dates=True,
sbert_summarizer=True)
elif args.method == 'clust_sbert':
cluster_ranker = clust.ClusterDateMentionCountRanker()
clusterer = clust.TemporalMarkovClusterer(max_days=365)
summarizer = summarizers.CentroidOpt()
system = clust.ClusteringTimelineGenerator(
clustering_rep='distilroberta-base-paraphrase-v1',
sbert_sequence_len=512,
cluster_ranker=cluster_ranker,
clusterer=clusterer,
summarizer=summarizer,
clip_sents=5,
unique_dates=True,
)
elif args.method == 'clust_sbert_sbertsum':
cluster_ranker = clust.ClusterDateMentionCountRanker()
clusterer = clust.TemporalMarkovClusterer(max_days=7)
summarizer = summarizers.SBERTSummarizer(date_only=True,
summary_criteria='similarity',
candidate_sents_per=5,
compare_with='both')
system = clust.ClusteringTimelineGenerator(
clustering_rep='distilroberta-base-paraphrase-v1',
sbert_sequence_len=512,
cluster_ranker=cluster_ranker,
clusterer=clusterer,
summarizer=summarizer,
sbert_summarizer=True,
clip_sents=5,
unique_dates=True,
)
elif args.method == 'clust_sbert_sbert_old':
cluster_ranker = clust.ClusterDateMentionCountRanker()
clusterer = clust.TemporalMarkovClusterer(max_days=7)
summarizer = summarizers.SubmodularSummarizer()
system = clust.ClusteringTimelineGenerator(
clustering_rep='distilroberta-base-paraphrase-v1',
sbert_sequence_len=512,
cluster_ranker=cluster_ranker,
clusterer=clusterer,
summarizer=summarizer,
summarizer_rep='same',
clip_sents=5,
unique_dates=True,
)
elif args.method == 'sbert':
system = clust.SBERTTimelineGenerator(
model_name='distilroberta-base-paraphrase-v1',
cd_n_articles=list(range(5, 20, 1)) + list(range(20, 100, 5)),
cd_thresholds=np.linspace(.25, .95, 25).tolist(),
cd_init_max_size=500,
min_comm_mult=1.5,
cluster_ranking='date_mention',
candidate_sents_per=5,
candidate_articles_per=10,
similarity_num_articles=10,
summary_criteria='similarity',
compare_with='both',
unique_dates=True)
# elif args.method == 'network':
# summarizer = summarizers.CentroidOpt()
# system = tr_network.NetworkTimelineGenerator()
else:
raise ValueError(f'Method not found: {args.method}')
if dataset_name == 'entities':
evaluate(system,
dataset,
args.output,
trunc_timelines=True,
time_span_extension=7,
word_mover_stop_words='nltk')
else:
evaluate(system,
dataset,
args.output,
trunc_timelines=False,
time_span_extension=0,
word_mover_stop_words='nltk')