def lda(args):
data = get_data(args)
mod = Aggregator(LDA, 1, data, args.size, categories=args.categories)
mod.train()
mod.get_metrics(data)
mod.mean(display_scores=True)
mod.std(display_scores=True)
def pca(args):
data = get_data(args)
mod = Aggregator(PCA, 1, data, args.size)
mod.train()
mod.get_metrics(data)
mod.mean(display_scores=True)
mod.std(display_scores=True)
def unsup_ae(args):
data = get_data(args)
mod = Aggregator(UnsupNN,
args.number,
data,
args.size,
reconstruct_loss=args.reconstruct_loss,
reconstruct_weight=args.reconstruct_weight,
enc_regularizer_weight=args.enc_regularizer_weight,
dec_regularizer_weight=args.dec_regularizer_weight,
lr=args.lr,
lr_decay=args.lr_decay,
encoder_regularizer=args.encoder_regularizer)
aggregated_keras(mod, data, args.model_path)
def sup_cats_ae(args):
data = get_data(args)
mod = Aggregator(SupNN,
args.number,
data,
args.size,
reconstruct_loss=args.reconstruct_loss,
reconstruct_weight=args.reconstruct_weight,
enc_regularizer_weight=args.enc_regularizer_weight,
dec_regularizer_weight=args.dec_regularizer_weight,
lr=args.lr,
lr_decay=args.lr_decay,
encoder_regularizer=args.encoder_regularizer)
aggregated_keras(mod, data, args.model_path)
mod.get_own_metrics(data, categories=True)
mod.mean(display_scores=True)
mod.std(display_scores=True)
def mds(args):
data = get_data(args)
mod = Aggregator(MDS,
args.number,
data,
args.size,
n_components=args.size,
metric=args.metric,
n_init=args.n_init,
max_iter=args.max_iter,
verbose=args.verbose,
eps=args.eps,
n_jobs=args.n_jobs,
random_state=args.random_state,
dissimilarity=args.dissimilarity)
aggregated(mod, data)
def tsne(args):
data = get_data(args)
mod = Aggregator(TSNE,
args.number,
data,
args.size,
perplexity=args.perplexity,
early_exaggeration=args.early_exaggeration,
learning_rate=args.learning_rate,
n_iter=args.n_iter,
n_iter_without_progress=args.n_iter_without_progress,
min_grad_norm=args.min_grad_norm,
metric=args.metric,
init=args.init,
verbose=args.verbose,
random_state=args.random_state,
method=args.method,
angle=args.angle)
_aggregated(mod, data, False, args.model_path)
def get_weights(args):
data = get_data(args)
weights1 = OrderedDict([('model', []), ('param', [])])
weights2 = OrderedDict([('model', []), ('param', [])])
for feat in data.columns:
weights1[feat] = []
weights2[feat] = []
# bin
if args.bin_ae_path is not None:
for w in args.bin_ae_weight:
model_path = args.bin_ae_path % w
assert os.path.exists(
model_path), 'model %s doesn\'t exist!' % model_path
agg = Aggregator(SemisupNN,
5,
data,
2,
categories=False,
reconstruct_loss='mse',
reconstruct_weight=float(w))
agg.load_models(model_path)
for mod in agg.models:
weights1['model'].append('bin_ae')
weights1['param'].append(w)
weights2['model'].append('bin_ae')
weights2['param'].append(w)
for name, (w1, w2) in zip(data.columns,
mod.get_feature_weights()):
weights1[name].append(w1)
weights2[name].append(w2)
# cats
if args.cats_ae_path is not None:
for w in args.cats_ae_weight:
model_path = args.cats_ae_path % w
assert os.path.exists(
model_path), 'model %s doesn\'t exist!' % model_path
agg = Aggregator(SemisupNN,
5,
data,
2,
categories=True,
reconstruct_loss='mse',
reconstruct_weight=float(w))
agg.load_models(model_path)
for mod in agg.models:
weights1['model'].append('cats_ae')
weights1['param'].append(w)
weights2['model'].append('cats_ae')
weights2['param'].append(w)
for name, (w1, w2) in zip(data.columns,
mod.get_feature_weights()):
weights1[name].append(w1)
weights2[name].append(w2)
print('Weights 1')
for k, v in weights1.items():
print(k, ','.join([str(vv) for vv in v]), sep=',')
print('Weights 2')
for k, v in weights2.items():
print(k, ','.join([str(vv) for vv in v]), sep=',')