def V_setup(self):
self.dim = self.input_data["input"].shape[1]
self.num_ob = self.input_data["target"].shape[0]
self.explicit_gradient = True
self.need_higherorderderiv = True
self.num_units = 10
prior_hidden_fn = prior_generator("gaussian_inv_gamma_2")
prior_out_fn = prior_generator("normal", var=1 / self.num_units)
self.hidden_in = prior_hidden_fn(obj=self,
name="hidden_in",
shape=(self.num_units, self.dim))
self.hidden_out = prior_out_fn(obj=self,
name="hidden_out",
shape=(2, self.num_units),
global_scale=1 / self.num_units)
#self.hidden_in_z = nn.Parameter(torch.zeros(self.num_units, self.dim), requires_grad=True)
#self.hidden_out_z = nn.Parameter(torch.zeros(2,self.num_units),requires_grad=True)
self.y = Variable(torch.from_numpy(self.y_np),
requires_grad=False).type("torch.LongTensor")
self.X = Variable(torch.from_numpy(self.X_np),
requires_grad=False).type(self.precision_type)
# include
self.dict_parameters = {
"hidden_in": self.hidden_in,
"hidden_out": self.hidden_out
}
return ()
def V_setup(self):
self.dim = self.input_data["input"].shape[1]
self.num_ob = self.input_data["target"].shape[0]
self.num_classes = len(numpy.unique(self.input_data["target"]))
self.explicit_gradient = True
self.need_higherorderderiv = True
self.num_units = self.model_dict["num_units"]
prior_hidden_fn = prior_generator(self.prior_dict["name"])
prior_out_fn = prior_generator("normal")
self.hidden_in = prior_hidden_fn(obj=self,
name="hidden_in",
shape=(self.num_units, self.dim),
global_scale=math.sqrt(1 / self.dim))
self.hidden_out = prior_out_fn(
obj=self,
name="hidden_out",
shape=(self.num_classes, self.num_units),
global_scale=math.sqrt(1 / self.num_units))
#self.hidden_in_z = nn.Parameter(torch.zeros(self.num_units, self.dim), requires_grad=True)
#self.hidden_out_z = nn.Parameter(torch.zeros(2,self.num_units),requires_grad=True)
self.y = Variable(torch.from_numpy(self.input_data["target"]),
requires_grad=False).type("torch.LongTensor")
self.X = Variable(torch.from_numpy(self.input_data["input"]),
requires_grad=False).type(self.precision_type)
self.dict_parameters = {
"hidden_in": self.hidden_in,
"hidden_out": self.hidden_out
}
# include
return ()
def V_setup(self):
self.dim = self.input_data["input"].shape[1]
self.num_ob = self.input_data["target"].shape[0]
self.num_classes = len(numpy.unique(self.input_data["target"]))
self.explicit_gradient = True
self.need_higherorderderiv = True
self.num_units = 35
self.num_layers = self.model_dict["num_layers"]
prior_fn = prior_generator("normal")
self.hidden_in = prior_fn(obj=self,
name="hidden_in",
shape=(self.num_units, self.dim),
global_scale=1)
self.hidden_out = prior_fn(obj=self,
name="hidden_out",
shape=(self.num_classes, self.num_units),
global_scale=1)
self.dict_parameters = {
"hidden_in": self.hidden_in,
"hidden_out": self.hidden_out
}
for i in range(self.num_layers - 1):
prior_hidden_fn = prior_generator(self.prior_dict["name"])
weights_name = "hidden_{}".format(i + 1)
obj = prior_hidden_fn(obj=self,
name=weights_name,
shape=(self.num_units, self.num_units),
global_scale=1)
setattr(self, weights_name, obj)
self.dict_parameters.update({weights_name: obj})
#self.hidden_in_z = nn.Parameter(torch.zeros(self.num_units, self.dim), requires_grad=True)
#self.hidden_out_z = nn.Parameter(torch.zeros(2,self.num_units),requires_grad=True)
self.y = Variable(torch.from_numpy(self.input_data["target"]),
requires_grad=False).type("torch.LongTensor")
self.X = Variable(torch.from_numpy(self.input_data["input"]),
requires_grad=False).type(self.precision_type)
# include
return ()
def V_setup(self):
self.dim = self.input_data["input"].shape[1]
self.num_ob = self.input_data["input"].shape[0]
self.explicit_gradient = False
self.need_higherorderderiv = True
prior_generator_fn = prior_generator(self.prior_dict["name"])
prior_obj = prior_generator_fn(obj=self, name="beta", shape=[self.dim])
# self.beta_obj = prior_obj.generator(obj=self,name="beta",shape=[self.dim])
self.beta_obj = prior_obj
self.y = Variable(torch.from_numpy(self.input_data["target"]), requires_grad=False).type(self.precision_type)
self.X = Variable(torch.from_numpy(self.input_data["input"]), requires_grad=False).type(self.precision_type)
self.dict_parameters = {"beta":self.beta_obj}
return ()
def V_setup(self):
self.dim = len(self.input_data["target"])
self.explicit_gradient = False
self.need_higherorderderiv = False
hs_prior_generator_fun = prior_generator("rhorseshoe_3")
prior_obj = hs_prior_generator_fun(obj=self,
name="beta",
shape=[self.dim])
self.beta_obj = prior_obj
self.y = Variable(torch.from_numpy(self.input_data["target"]),
requires_grad=False).type(self.precision_type)
self.dict_parameters = {"beta": self.beta_obj}
return ()
def V_setup(self):
self.dim = self.X_np.shape[1]
self.num_ob = self.X_np.shape[0]
self.explicit_gradient = False
self.need_higherorderderiv = True
prior_generator_fn = prior_generator("horseshoe_1")
prior_obj = prior_generator_fn(obj=self, name="beta", shape=[self.dim])
#self.beta_obj = prior_obj.generator(obj=self,name="beta",shape=[self.dim])
self.beta_obj = prior_obj
self.y = Variable(torch.from_numpy(self.input_data["target"]),
requires_grad=False).type(self.precision_type)
self.X = Variable(torch.from_numpy(self.input_data["input"]),
requires_grad=False).type(self.precision_type)
# include
#self.sigma =1
return ()
