input_sparsity = 0.1
w_sparsity = 0.1
input_scaling = 0.5
n_test = 20
n_samples = 5000
leaky_rate = 0.1
reservoir_size = 400
# Argument
args = tools.functions.argument_parser_training_model()
# Transformer
if args.lang == 'en' or args.lang == 'fr':
transformer = transforms.Compose([
# transforms.RemoveLines(),
transforms.GloveVector(
model=tools.settings.lang_models[socket.gethostname()][args.lang])
])
else:
transformer = transforms.Compose([
# transforms.RemoveLines(),
transforms.Token(model=tools.settings.lang_spacy_models[args.lang],
lang=tools.settings.lang_models_lang[args.lang]),
transforms.GensimModel(model_path=tools.settings.lang_models[
socket.gethostname()][args.lang])
])
# end if
# Samples average
samples_average = np.array([])
best_acc = 0
best_model = None
args.description,
args.get_space(),
args.n_samples,
args.k,
verbose=args.verbose
)
# CNN Glove Feature Selector
cgfs = models.cgfs(pretrained=True, n_gram=2, n_features=60)
# Remove last linear layer
cgfs.linear2 = echotorch.nn.Identity()
# Transformer
transformer = transforms.Compose([
transforms.GloveVector(),
transforms.ToNGram(n=2, overlapse=True),
transforms.Reshape((-1, 1, 2, 300)),
transforms.FeatureSelector(cgfs, 60, to_variable=True),
transforms.Reshape((1, -1, 60)),
transforms.Normalize(mean=-4.56512329954, std=0.911449706065)
])
# Reuters C50 dataset
reutersloader = torch.utils.data.DataLoader(datasets.ReutersC50Dataset(
root=args.dataset,
download=True,
n_authors=args.n_authors,
transform=transformer),
batch_size=1,
shuffle=False)
# Experiment settings
spectral_radius = 0.95
input_sparsity = 0.1
w_sparsity = 0.1
input_scaling = 0.5
n_test = 20
n_samples = 30
leaky_rate = 0.1
# Argument
args = tools.functions.argument_parser_training_model()
# Transformer
transformer = transforms.Compose([
transforms.RemoveLines(),
transforms.GloveVector(model=tools.settings.lang_models[args.lang])
])
# Results
parameter_averages = np.zeros(n_test)
parameter_max = np.zeros(n_test)
# For each leaky rate values
index = 0
for rc_reservoir_size in np.linspace(200, 1000, n_test):
# Round
reservoir_size = int(math.floor(rc_reservoir_size))
# Log
print(u"Reservoir size : {}".format(reservoir_size))