libcity.model.traffic_speed_prediction.GTS

class libcity.model.traffic_speed_prediction.GTS.DCGRUCell(input_dim, num_units, adj_mx, max_diffusion_step, num_nodes, device, nonlinearity='tanh', filter_type='laplacian', use_gc_for_ru=True)

Bases: torch.nn.modules.module.Module

forward(inputs, hx)

Gated recurrent unit (GRU) with Graph Convolution. :param inputs: (B, num_nodes * input_dim) :param hx: (B, num_nodes * rnn_units)

Returns

shape (B, num_nodes * rnn_units)

Return type

torch.tensor

training: bool

class libcity.model.traffic_speed_prediction.GTS.DecoderModel(config, data_feature, adj_mx, device)

Bases: torch.nn.modules.module.Module, libcity.model.traffic_speed_prediction.GTS.Seq2SeqAttrs

forward(inputs, hidden_state=None)

Decoder forward pass. :param inputs: shape (batch_size, self.num_nodes * self.output_dim) :param hidden_state: (num_layers, batch_size, self.hidden_state_size),

optional, zeros if not provided, hidden_state_size = num_nodes * rnn_units

Returns

tuple contains:

output: shape (batch_size, self.num_nodes * self.output_dim)

hidden_state: shape (num_layers, batch_size, self.hidden_state_size)

(lower indices mean lower layers)

Return type

tuple

training: bool

class libcity.model.traffic_speed_prediction.GTS.EncoderModel(config, data_feature, adj_mx, device)

Bases: torch.nn.modules.module.Module, libcity.model.traffic_speed_prediction.GTS.Seq2SeqAttrs

forward(inputs, hidden_state=None)

Encoder forward pass. :param inputs: shape (batch_size, self.num_nodes * self.input_dim) :param hidden_state: (num_layers, batch_size, self.hidden_state_size),

optional, zeros if not provided, hidden_state_size = num_nodes * rnn_units

Returns

tuple contains:

output: shape (batch_size, self.hidden_state_size)

hidden_state: shape (num_layers, batch_size, self.hidden_state_size)

(lower indices mean lower layers)

Return type

tuple

training: bool

class libcity.model.traffic_speed_prediction.GTS.FC(num_nodes, device, input_dim, hid_dim, output_dim, bias_start=0.0)

Bases: torch.nn.modules.module.Module

forward(inputs, state)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class libcity.model.traffic_speed_prediction.GTS.GCONV(num_nodes, max_diffusion_step, device, input_dim, hid_dim, output_dim, adj_mx, bias_start=0.0)

Bases: torch.nn.modules.module.Module

forward(inputs, state)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class libcity.model.traffic_speed_prediction.GTS.GTS(config, data_feature)

Bases: libcity.model.abstract_traffic_state_model.AbstractTrafficStateModel, libcity.model.traffic_speed_prediction.GTS.Seq2SeqAttrs

_get_x_y(x, y)

Parameters

• x – shape (batch_size, seq_len, num_sensor, input_dim)

• y – shape (batch_size, horizon, num_sensor, input_dim)

:returns x shape (seq_len, batch_size, num_sensor, input_dim)

y shape (horizon, batch_size, num_sensor, input_dim)

_get_x_y_in_correct_dims(x, y)

Parameters

• x – shape (seq_len, batch_size, num_sensor, input_dim)

• y – shape (horizon, batch_size, num_sensor, input_dim)

Returns

x: shape (seq_len, batch_size, num_sensor * input_dim) y: shape (horizon, batch_size, num_sensor * output_dim)

calculate_loss(batch, batches_seen=None)

输入一个batch的数据，返回训练过程的loss，也就是需要定义一个loss函数

Parameters

batch (Batch) – a batch of input

Returns

return training loss

Return type

torch.tensor

decoder(encoder_hidden_state, labels=None, batches_seen=None)

Decoder forward pass :param encoder_hidden_state: (num_layers, batch_size, self.hidden_state_size) :param labels: (self.horizon, batch_size, self.num_nodes * self.output_dim) [optional, not exist for inference] :param batches_seen: global step [optional, not exist for inference] :return: output: (self.horizon, batch_size, self.num_nodes * self.output_dim)

encoder(inputs)

Encoder forward pass :param inputs: shape (seq_len, batch_size, num_sensor * input_dim) :return: encoder_hidden_state: (num_layers, batch_size, self.hidden_state_size)

forward(batch, batches_seen=None)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

predict(batch, batches_seen=None)

输入一个batch的数据，返回对应的预测值，一般应该是**多步预测**的结果，一般会调用nn.Moudle的forward()方法

Parameters

batch (Batch) – a batch of input

Returns

predict result of this batch

Return type

torch.tensor

training: bool

class libcity.model.traffic_speed_prediction.GTS.Seq2SeqAttrs(config, data_feature)

Bases: object

libcity.model.traffic_speed_prediction.GTS.cosine_similarity_torch(x1, x2=None, eps=1e-08)

libcity.model.traffic_speed_prediction.GTS.count_parameters(model)

libcity.model.traffic_speed_prediction.GTS.gumbel_softmax(device, logits, temperature, hard=False, eps=1e-10)

Sample from the Gumbel-Softmax distribution and optionally discretize. :param logits: [batch_size, n_class] unnormalized log-probs :param temperature: non-negative scalar :param hard: if True, take argmax, but differentiate w.r.t. soft sample y

Returns

[batch_size, n_class] sample from the Gumbel-Softmax distribution. If hard=True, then the returned sample will be one-hot, otherwise it will be a probabilitiy distribution that sums to 1 across classes

libcity.model.traffic_speed_prediction.GTS.gumbel_softmax_sample(device, logits, temperature, eps=1e-10)

libcity.model.traffic_speed_prediction.GTS.sample_gumbel(device, shape, eps=1e-20)