def train(self,
embedding_size=128,
window_size=5,
workers=8,
nb_epochs=50,
**kwargs):
kwargs["sentences"] = self.walks
kwargs["min_count"] = kwargs.get("min_count", 0)
kwargs["size"] = embedding_size
kwargs["sg"] = 1 # skip gram
kwargs["hs"] = 1 # deepwalk use Hierarchical Softmax
kwargs["workers"] = workers
kwargs["window"] = window_size
kwargs["iter"] = nb_epochs
kwargs["compute_loss"] = True
kwargs["min_alpha"] = 1e-2
call_back = callback(all_loss=[])
kwargs['callbacks'] = [call_back]
print("Learning embedding vectors...")
self.word_vectors = Word2Vec(**kwargs).wv
print("Learning embedding vectors done!")
self.word_vectors = dict(
zip(self.word_vectors.index2word,
self.word_vectors.vectors.tolist()))
verbose='OFF')
if args.generator != 'ON':
train_sample = apply_t_scaler(train_sample, t_scaler, verbose='ON')
sample_composition(train_sample)
compo_matrix(valid_labels, train_labels=train_labels)
print()
train_weights, bins = get_sample_weights(train_sample,
train_labels,
args.weight_type,
args.bkg_ratio,
hist='pt')
sample_histograms(valid_sample, valid_labels, train_sample, train_labels,
train_weights, bins, args.output_dir)
#sys.exit()
callbacks = callback(args.model_out, args.patience, args.metrics)
if args.generator == 'ON':
del (train_sample)
if np.all(train_weights) != None:
train_weights = gen_weights(args.n_train, weight_idx,
train_weights)
print('\nLAUNCHING GENERATOR FOR',
np.diff(args.n_train)[0], 'TRAINING SAMPLES')
eval_gen = Batch_Generator(data_files,
args.n_eval,
input_data,
args.n_tracks,
args.n_classes,
valid_batch_size,
args.valid_cuts,
scaler,
s_scaler)
def separate_samples(sample):
JZW = sample['JZW']
qcd_sample = {key: val[JZW != -1] for key, val in sample.items()}
oe_sample = {key: val[JZW == -1] for key, val in sample.items()}
return qcd_sample, oe_sample
train_sample, oe_train_sample = separate_samples(train_sample)
valid_sample, oe_valid_sample = separate_samples(valid_sample)
print('\nTrain sample:', format(len(train_sample['weights']), '8.0f'),
'jets')
print('Valid sample:', format(len(valid_sample['weights']), '8.0f'),
'jets\n')
print('Using TensorFlow', tf.__version__, 'with', n_gpus, 'GPU(s)\n')
callbacks = callback(args.model_out, args.patience, 'val_loss')
#FOR STAND-ALONE ENCODER
if args.encoder in ['ae', 'vae']:
sample_weights = train_sample['weights']
if len(set(train_var) - {'constituents'}) == 0:
train_X = train_sample['constituents']
valid_X = valid_sample['constituents']
elif 'constituents' not in train_var:
train_X = train_sample['scalars']
valid_X = valid_sample['scalars']
else:
train_X = np.concatenate(
[train_sample['constituents'], train_sample['scalars']],
axis=1)
valid_X = np.concatenate(
[valid_sample['constituents'], valid_sample['scalars']],