def test_simple_mlp_creating():
X = cn.ParameterInput([20, 784], "X")
Y = cn.ParameterInput([20, 10], "Y")
hl = cn.mlp_layer(X, 100, cn.mlp_layer_opts().tanh().group("hl"))
lr = cn.logistic_regression(hl.output, Y, False)
with open("bla.dot", "w") as dotfile:
dotfile.write(lr.loss.dot(True))
dw = xdot.DotWindow()
dw.connect('destroy', gtk.main_quit)
dw.set_filter('dot')
dw.set_dotcode(lr.loss.dot(True))
gtk.main()
def __init__(self, args):
self.args = args
cfg = cn.mlp_layer_opts()
for k, v in IterNotNone(args):
if k == "size":
self.size = int(v)
elif k == "group":
if isinstance(v, tuple):
cfg.group(*v)
else:
cfg.group(v)
elif k == "verbose":
cfg.verbose(v)
elif k == "dropout":
cfg.dropout(v)
elif k == "nonlin":
if v == "linear":
pass
elif v == "rectified_linear":
cfg.rectified_linear()
elif v == "tanh":
cfg.tanh()
else:
raise RuntimeError("Unknown non-linearity type `%s'" % str(nonlin))
elif k == "n_groups":
cfg.n_groups(v)
elif k == "maxout":
cfg.maxout(v)
elif k == "lr_fact":
cfg.learnrate_factor(v)
elif k == "with_bias":
if isinstance(v, tuple):
cfg.with_bias(*v)
else:
cfg.with_bias(v)
elif k == "init_std":
cfg.weight_init_std(v)
else:
raise RuntimeError("Unknown argument `%s'" % k)
self.cfg = cfg