# Copyright 2024 All authors of TrajDL
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License 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 warnings
from typing import List
import torch
from torch import nn
from torch.nn.utils.rnn import pack_padded_sequence, pad_packed_sequence
[docs]
class RNNNextLocPredictor(nn.Module):
def __init__(
self,
embedding_layer: nn.Embedding,
rnn_hidden_size: int,
fc_hidden_size: int,
output_size: int,
num_layers: int,
padding_value: int,
dropout: float = 0.0,
):
super().__init__()
self.embedding_layer = embedding_layer
self.add_module("embedding_layer", self.embedding_layer)
self.rnn = nn.RNN(
embedding_layer.embedding_dim,
rnn_hidden_size,
num_layers,
dropout=dropout,
batch_first=True,
)
self.out_linear = nn.Sequential(
nn.Tanh(),
nn.Linear(rnn_hidden_size, fc_hidden_size),
nn.LeakyReLU(),
nn.Linear(fc_hidden_size, output_size),
)
self._pad = padding_value
[docs]
def forward(self, src: torch.LongTensor, lengths: List[int]):
"""
Parameters
----------
src: shape is (B, T)
lengths: List[int]
"""
# (B, T, C)
emb = self.embedding_layer(src)
inputs = pack_padded_sequence(
input=emb, lengths=lengths, batch_first=True, enforce_sorted=False
)
# (B, T, H), (B, H)
output, _ = self.rnn(inputs)
# (B, T, H), (B,)
output, _ = pad_packed_sequence(
output, batch_first=True, padding_value=self._pad
)
return output
[docs]
def select_last_k_elements(x: torch.Tensor, lengths: List[int], k: int) -> torch.Tensor:
"""
้ๅบๅบๅ้้ขๆๅkไธชๅ
็ด ๏ผ่ฟ้่ฆ่่ๆฏไธชๅบๅ็ๅฎ้
้ฟๅบฆ๏ผxๆฏpaddingๅ็ๅบๅ
shape of x is (B, T, *)
"""
if min(lengths) < 0:
raise ValueError(
f"The minimum of lengths should be a positive number or 0, not {min(lengths)}"
)
if min(lengths) == 0:
warnings.warn("There is a 0 value in `lengths` parameter", RuntimeWarning)
first_indices = torch.arange(x.shape[0]).unsqueeze(dim=1)
second_indices = torch.LongTensor(lengths).unsqueeze(dim=1) - k + torch.arange(k)
# (B, k, *)
return x[first_indices, second_indices]
