def __init__(
self,
input_size: int,
num_layers: int,
num_cells: int,
cell_type: str,
history_length: int,
context_length: int,
prediction_length: int,
distr_output: DistributionOutput,
control_output: DistributionOutput,
dropout_rate: float,
cardinality: List[int],
embedding_dimension: List[int],
lags_seq: List[int],
scaling: bool = True,
dtype: np.dtype = np.float32,
) -> None:
super().__init__()
self.num_layers = num_layers
self.num_cells = num_cells
self.cell_type = cell_type
self.history_length = history_length
self.context_length = context_length
self.prediction_length = prediction_length
self.dropout_rate = dropout_rate
self.cardinality = cardinality
self.embedding_dimension = embedding_dimension
self.num_cat = len(cardinality)
self.scaling = scaling
self.dtype = dtype
self.lags_seq = lags_seq
self.distr_output = distr_output
self.control_output = control_output
rnn = {"LSTM": nn.LSTM, "GRU": nn.GRU}[self.cell_type]
self.rnn = rnn(
input_size=input_size,
hidden_size=num_cells,
num_layers=num_layers,
dropout=dropout_rate,
batch_first=True,
)
self.target_shape = distr_output.event_shape
self.proj_distr_args = distr_output.get_args_proj(num_cells + 1)
self.proj_control_args = control_output.get_args_proj(num_cells)
self.embedder = FeatureEmbedder(cardinalities=cardinality,
embedding_dims=embedding_dimension)
if scaling:
self.scaler = MeanScaler(keepdim=True)
self.control_scaler = NOPScaler(keepdim=True)
else:
self.scaler = NOPScaler(keepdim=True)
self.control_scaler = NOPScaler(keepdim=True)
def __init__(
self,
num_hidden_dimensions: List[int],
prediction_length: int,
context_length: int,
batch_normalization: bool,
mean_scaling: bool,
distr_output: DistributionOutput,
) -> None:
super().__init__()
self.num_hidden_dimensions = num_hidden_dimensions
self.prediction_length = prediction_length
self.context_length = context_length
self.batch_normalization = batch_normalization
self.mean_scaling = mean_scaling
self.distr_output = distr_output
modules = []
dims = self.num_hidden_dimensions
for i, units in enumerate(dims[:-1]):
if i == 0:
input_size = context_length
else:
input_size = dims[i - 1]
modules += [nn.Linear(input_size, units), nn.ReLU()]
if self.batch_normalization:
modules.append(nn.BatchNorm1d(units))
if len(dims) == 1:
modules.append(
nn.Linear(context_length, dims[-1] * prediction_length))
else:
modules.append(nn.Linear(dims[-2], dims[-1] * prediction_length))
modules.append(
LambdaLayer(
lambda o: torch.reshape(o, (-1, prediction_length, dims[-1]))))
self.mlp = nn.Sequential(*modules)
self.distr_args_proj = self.distr_output.get_args_proj(dims[-1])
self.criterion = nn.SmoothL1Loss(reduction="none")
self.scaler = MeanScaler() if mean_scaling else NOPScaler()
def __init__(
self,
input_size: int,
d_model: int,
num_heads: int,
act_type: str,
dropout_rate: float,
dim_feedforward_scale: int,
num_encoder_layers: int,
num_decoder_layers: int,
history_length: int,
context_length: int,
prediction_length: int,
lags_seq: List[int],
target_dim: int,
conditioning_length: int,
flow_type: str,
n_blocks: int,
hidden_size: int,
n_hidden: int,
dequantize: bool,
cardinality: List[int] = [1],
embedding_dimension: int = 1,
scaling: bool = True,
**kwargs,
) -> None:
super().__init__(**kwargs)
self.target_dim = target_dim
self.prediction_length = prediction_length
self.context_length = context_length
self.history_length = history_length
self.scaling = scaling
assert len(
set(lags_seq)) == len(lags_seq), "no duplicated lags allowed!"
lags_seq.sort()
self.lags_seq = lags_seq
self.encoder_input = nn.Linear(input_size, d_model)
self.decoder_input = nn.Linear(input_size, d_model)
# [B, T, d_model] where d_model / num_heads is int
self.transformer = nn.Transformer(
d_model=d_model,
nhead=num_heads,
num_encoder_layers=num_encoder_layers,
num_decoder_layers=num_decoder_layers,
dim_feedforward=dim_feedforward_scale * d_model,
dropout=dropout_rate,
activation=act_type,
)
flow_cls = {
"RealNVP": RealNVP,
"MAF": MAF,
}[flow_type]
self.flow = flow_cls(
input_size=target_dim,
n_blocks=n_blocks,
n_hidden=n_hidden,
hidden_size=hidden_size,
cond_label_size=conditioning_length,
)
self.dequantize = dequantize
self.distr_output = FlowOutput(self.flow,
input_size=target_dim,
cond_size=conditioning_length)
self.proj_dist_args = self.distr_output.get_args_proj(d_model)
self.embed_dim = 1
self.embed = nn.Embedding(num_embeddings=self.target_dim,
embedding_dim=self.embed_dim)
if self.scaling:
self.scaler = MeanScaler(keepdim=True)
else:
self.scaler = NOPScaler(keepdim=True)
# mask
self.register_buffer(
"tgt_mask",
self.transformer.generate_square_subsequent_mask(
prediction_length),
)
def __init__(
self,
input_size: int,
num_layers: int,
num_cells: int,
cell_type: str,
history_length: int,
context_length: int,
prediction_length: int,
distr_output: DistributionOutput,
dropout_rate: float,
lags_seq: List[int],
target_dim: int,
conditioning_length: int,
cardinality: List[int] = [1],
embedding_dimension: int = 1,
scaling: bool = True,
**kwargs,
) -> None:
super().__init__(**kwargs)
self.num_layers = num_layers
self.num_cells = num_cells
self.cell_type = cell_type
self.history_length = history_length
self.context_length = context_length
self.prediction_length = prediction_length
self.dropout_rate = dropout_rate
self.cardinality = cardinality
self.embedding_dimension = embedding_dimension
self.num_cat = len(cardinality)
self.target_dim = target_dim
self.scaling = scaling
self.target_dim_sample = target_dim
self.conditioning_length = conditioning_length
assert len(
set(lags_seq)) == len(lags_seq), "no duplicated lags allowed!"
lags_seq.sort()
self.lags_seq = lags_seq
self.distr_output = distr_output
self.target_dim = target_dim
rnn = {"LSTM": nn.LSTM, "GRU": nn.GRU}[self.cell_type]
self.rnn = rnn(
input_size=input_size,
hidden_size=num_cells,
num_layers=num_layers,
dropout=dropout_rate,
batch_first=True,
)
self.proj_dist_args = distr_output.get_args_proj(num_cells)
self.embed_dim = 1
self.embed = nn.Embedding(num_embeddings=self.target_dim,
embedding_dim=self.embed_dim)
if scaling:
self.scaler = MeanScaler(keepdim=True)
else:
self.scaler = NOPScaler(keepdim=True)
def __init__(
self,
input_size: int,
decoder_size: int,
num_layers: int,
num_cells: int,
cell_type: str,
short_cycle: int,
history_length: int,
context_length: int, # should be equal to the length of long period
prediction_length: int, # should be integer multiples of small period
distr_output: DistributionOutput,
dropout_rate: float,
cardinality: List[int],
embedding_dimension: List[int],
lags_seq: List[int],
scaling: bool = True,
dtype: np.dtype = np.float32,
) -> None:
super().__init__()
self.num_layers = num_layers
self.num_cells = num_cells
self.cell_type = cell_type
self.short_cycle = short_cycle
self.history_length = history_length
self.context_length = context_length
self.prediction_length = prediction_length
self.dropout_rate = dropout_rate
self.cardinality = cardinality
self.embedding_dimension = embedding_dimension
self.num_cat = len(cardinality)
self.scaling = scaling
self.dtype = dtype
# self.lags_seq = lags_seq + [0] # 0 is the current value
self.lags_seq = [l - 1 for l in lags_seq] # to distinguish with deepAR
self.distr_output = distr_output
rnn = {"LSTM": nn.LSTM, "GRU": nn.GRU}[self.cell_type]
self.long_cell = nn.GRUCell(input_size=input_size, hidden_size=num_cells)
self.encoder = nn.ModuleList(
rnn(
input_size=num_cells,
hidden_size=num_cells,
num_layers=num_layers,
dropout=dropout_rate,
batch_first=True,
)
for _ in range(self.context_length // self.short_cycle)
)
# self.encoder = rnn(
# input_size=num_cells,
# hidden_size=num_cells,
# num_layers=num_layers,
# dropout=dropout_rate,
# batch_first=True,
# )
self.decoder = rnn(
input_size=decoder_size,
hidden_size=num_cells,
num_layers=num_layers,
dropout=dropout_rate,
batch_first=True,
)
self.decoder_dnn = nn.ModuleList(nn.Sequential(
nn.Linear(decoder_size, 1),
# nn.ReLU()
) for _ in range(self.prediction_length))
self.out = nn.Linear(num_cells, 1)
self.criterion = nn.MSELoss() # l2, l1
self.target_shape = distr_output.event_shape
self.proj_distr_args = distr_output.get_args_proj(num_cells)
self.embedder = FeatureEmbedder(
cardinalities=cardinality, embedding_dims=embedding_dimension
)
if scaling:
self.scaler = MeanScaler(keepdim=True)
else:
self.scaler = NOPScaler(keepdim=True)
def __init__(
self,
input_size: int,
num_layers: int,
num_cells: int,
cell_type: str,
history_length: int,
context_length: int,
prediction_length: int,
dropout_rate: float,
lags_seq: List[int],
target_dim: int,
conditioning_length: int,
flow_type: str,
n_blocks: int,
hidden_size: int,
n_hidden: int,
dequantize: bool,
cardinality: List[int] = [1],
embedding_dimension: int = 1,
scaling: bool = True,
**kwargs,
) -> None:
super().__init__(**kwargs)
self.target_dim = target_dim
self.prediction_length = prediction_length
self.context_length = context_length
self.history_length = history_length
self.scaling = scaling
assert len(
set(lags_seq)) == len(lags_seq), "no duplicated lags allowed!"
lags_seq.sort()
self.lags_seq = lags_seq
self.cell_type = cell_type
rnn_cls = {"LSTM": nn.LSTM, "GRU": nn.GRU}[cell_type]
self.rnn = rnn_cls(
input_size=input_size,
hidden_size=num_cells,
num_layers=num_layers,
dropout=dropout_rate,
batch_first=True,
)
flow_cls = {
"RealNVP": RealNVP,
"MAF": MAF,
}[flow_type]
self.flow = flow_cls(
input_size=target_dim,
n_blocks=n_blocks,
n_hidden=n_hidden,
hidden_size=hidden_size,
cond_label_size=conditioning_length,
)
self.dequantize = dequantize
self.distr_output = FlowOutput(self.flow,
input_size=target_dim,
cond_size=conditioning_length)
self.proj_dist_args = self.distr_output.get_args_proj(num_cells)
self.embed_dim = 1
self.embed = nn.Embedding(num_embeddings=self.target_dim,
embedding_dim=self.embed_dim)
if self.scaling:
self.scaler = MeanScaler(keepdim=True)
else:
self.scaler = NOPScaler(keepdim=True)
# http://www.apache.org/licenses/LICENSE-2.0
#
# or in the "license" file accompanying this file. This file is distributed
# on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either
# express or implied. See the License for the specific language governing
# permissions and limitations under the License.
import numpy as np
import pytest
import torch
from pts.modules import MeanScaler, NOPScaler
test_cases = [
(
MeanScaler(),
torch.tensor([
[1.0] * 50,
[0.0] * 25 + [3.0] * 25,
[2.0] * 49 + [1.5] * 1,
[0.0] * 50,
[1.0] * 50,
]),
torch.tensor([
[1.0] * 50,
[0.0] * 25 + [1.0] * 25,
[0.0] * 49 + [1.0] * 1,
[1.0] * 50,
[0.0] * 50,
]),
torch.tensor([1.0, 3.0, 1.5, 1.00396824, 1.00396824]),
def __init__(
self,
input_size: int,
d_model: int,
num_heads: int,
act_type: str,
dropout_rate: float,
dim_feedforward_scale: int,
num_encoder_layers: int,
num_decoder_layers: int,
history_length: int,
context_length: int,
prediction_length: int,
distr_output: DistributionOutput,
cardinality: List[int],
embedding_dimension: List[int],
lags_seq: List[int],
scaling: bool = True,
**kwargs,
) -> None:
super().__init__(**kwargs)
self.history_length = history_length
self.context_length = context_length
self.prediction_length = prediction_length
self.scaling = scaling
self.cardinality = cardinality
self.embedding_dimension = embedding_dimension
self.distr_output = distr_output
assert len(set(lags_seq)) == len(lags_seq), "no duplicated lags allowed!"
lags_seq.sort()
self.lags_seq = lags_seq
self.target_shape = distr_output.event_shape
# [B, T, input_size] -> [B, T, d_model]
self.encoder_input = nn.Linear(input_size, d_model)
self.decoder_input = nn.Linear(input_size, d_model)
# [B, T, d_model] where d_model / num_heads is int
self.transformer = nn.Transformer(
d_model=d_model,
nhead=num_heads,
num_encoder_layers=num_encoder_layers,
num_decoder_layers=num_decoder_layers,
dim_feedforward=dim_feedforward_scale * d_model,
dropout=dropout_rate,
activation=act_type,
)
self.proj_dist_args = distr_output.get_args_proj(d_model)
self.embedder = FeatureEmbedder(
cardinalities=cardinality, embedding_dims=embedding_dimension,
)
if scaling:
self.scaler = MeanScaler(keepdim=True)
else:
self.scaler = NOPScaler(keepdim=True)
# mask
self.register_buffer(
"tgt_mask", self.transformer.generate_square_subsequent_mask(prediction_length)
)
def __init__(
self,
num_series: int,
channels: int,
kernel_size: int,
rnn_cell_type: str,
rnn_num_cells: int,
skip_rnn_cell_type: str,
skip_rnn_num_cells: int,
skip_size: int,
ar_window: int,
context_length: int,
horizon: Optional[int],
prediction_length: Optional[int],
dropout_rate: float,
output_activation: Optional[str],
scaling: bool,
*args,
**kwargs,
) -> None:
super().__init__(*args, **kwargs)
self.num_series = num_series
self.channels = channels
assert (
channels % skip_size == 0
), "number of conv1d `channels` must be divisible by the `skip_size`"
self.skip_size = skip_size
assert ar_window > 0, "auto-regressive window must be a positive integer"
self.ar_window = ar_window
assert not ((horizon is None)) == (
prediction_length is None
), "Exactly one of `horizon` and `prediction_length` must be set at a time"
assert horizon is None or horizon > 0, "`horizon` must be greater than zero"
assert (prediction_length is None or prediction_length > 0
), "`prediction_length` must be greater than zero"
self.prediction_length = prediction_length
self.horizon = horizon
assert context_length > 0, "`context_length` must be greater than zero"
self.context_length = context_length
if output_activation is not None:
assert output_activation in [
"sigmoid",
"tanh",
], "`output_activation` must be either 'sigmiod' or 'tanh' "
self.output_activation = output_activation
assert rnn_cell_type in [
"GRU",
"LSTM",
], "`rnn_cell_type` must be either 'GRU' or 'LSTM' "
assert skip_rnn_cell_type in [
"GRU",
"LSTM",
], "`skip_rnn_cell_type` must be either 'GRU' or 'LSTM' "
conv_out = context_length - kernel_size
self.conv_skip = conv_out // skip_size
assert self.conv_skip > 0, (
"conv1d output size must be greater than or equal to `skip_size`\n"
"Choose a smaller `kernel_size` or bigger `context_length`")
self.cnn = nn.Conv2d(in_channels=1,
out_channels=channels,
kernel_size=(num_series, kernel_size))
self.dropout = nn.Dropout(p=dropout_rate)
rnn = {"LSTM": nn.LSTM, "GRU": nn.GRU}[rnn_cell_type]
self.rnn = rnn(
input_size=channels,
hidden_size=rnn_num_cells,
# dropout=dropout_rate,
)
skip_rnn = {"LSTM": nn.LSTM, "GRU": nn.GRU}[skip_rnn_cell_type]
self.skip_rnn_num_cells = skip_rnn_num_cells
self.skip_rnn = skip_rnn(
input_size=channels,
hidden_size=skip_rnn_num_cells,
# dropout=dropout_rate,
)
self.fc = nn.Linear(rnn_num_cells + skip_size * skip_rnn_num_cells,
num_series)
if self.horizon:
self.ar_fc = nn.Linear(ar_window, 1)
else:
self.ar_fc = nn.Linear(ar_window, prediction_length)
if scaling:
self.scaler = MeanScaler(keepdim=True, time_first=False)
else:
self.scaler = NOPScaler(keepdim=True, time_first=False)
def __init__(
self,
input_size: int,
num_layers: int,
num_cells: int,
cell_type: str,
history_length: int,
context_length: int,
prediction_length: int,
dropout_rate: float,
lags_seq: List[int],
target_dim: int,
conditioning_length: int,
diff_steps: int,
loss_type: str,
beta_end: float,
beta_schedule: str,
residual_layers: int,
residual_channels: int,
dilation_cycle_length: int,
cardinality: List[int] = [1],
embedding_dimension: int = 1,
scaling: bool = True,
**kwargs,
) -> None:
super().__init__(**kwargs)
self.target_dim = target_dim
self.prediction_length = prediction_length
self.context_length = context_length
self.history_length = history_length
self.scaling = scaling
assert len(
set(lags_seq)) == len(lags_seq), "no duplicated lags allowed!"
lags_seq.sort()
self.lags_seq = lags_seq
self.cell_type = cell_type
rnn_cls = {"LSTM": nn.LSTM, "GRU": nn.GRU}[cell_type]
self.rnn = rnn_cls(
input_size=input_size,
hidden_size=num_cells,
num_layers=num_layers,
dropout=dropout_rate,
batch_first=True,
)
self.denoise_fn = EpsilonTheta(
target_dim=target_dim,
cond_length=conditioning_length,
residual_layers=residual_layers,
residual_channels=residual_channels,
dilation_cycle_length=dilation_cycle_length,
)
self.diffusion = GaussianDiffusion(
self.denoise_fn,
input_size=target_dim,
diff_steps=diff_steps,
loss_type=loss_type,
beta_end=beta_end,
beta_schedule=beta_schedule,
)
self.distr_output = DiffusionOutput(self.diffusion,
input_size=target_dim,
cond_size=conditioning_length)
self.proj_dist_args = self.distr_output.get_args_proj(num_cells)
self.embed_dim = 1
self.embed = nn.Embedding(num_embeddings=self.target_dim,
embedding_dim=self.embed_dim)
if self.scaling:
self.scaler = MeanScaler(keepdim=True)
else:
self.scaler = NOPScaler(keepdim=True)