(self, input_batch: CommitClassifierInputs)
| 94 | self.activation_fn = config.activation() |
| 95 | |
| 96 | def forward(self, input_batch: CommitClassifierInputs): |
| 97 | # Encode input title |
| 98 | title: List[str] = input_batch.title |
| 99 | model_input = to_tensor(self.transform(title), padding_value=1).to(device) |
| 100 | title_features = self.encoder(model_input) |
| 101 | title_embed = title_features[:, 0, :] |
| 102 | title_embed = self.dropout(title_embed) |
| 103 | title_embed = self.dense_title(title_embed) |
| 104 | title_embed = self.activation_fn(title_embed) |
| 105 | title_embed = self.dropout(title_embed) |
| 106 | title_embed = self.out_proj_title(title_embed) |
| 107 | |
| 108 | files: list[str] = input_batch.files |
| 109 | batch_file_indexes = [] |
| 110 | for file in files: |
| 111 | paths = [ |
| 112 | truncate_file(Path(file_part), MAX_LEN_FILE) |
| 113 | for file_part in file.split(" ") |
| 114 | ] |
| 115 | batch_file_indexes.append( |
| 116 | [ |
| 117 | self.file_map.get(file, self.file_map[UNKNOWN_TOKEN]) |
| 118 | for file in paths |
| 119 | ] |
| 120 | ) |
| 121 | |
| 122 | flat_indexes = torch.tensor( |
| 123 | list(chain.from_iterable(batch_file_indexes)), |
| 124 | dtype=torch.long, |
| 125 | device=device, |
| 126 | ) |
| 127 | offsets = [0] |
| 128 | offsets.extend(len(files) for files in batch_file_indexes[:-1]) |
| 129 | offsets = torch.tensor(offsets, dtype=torch.long, device=device) |
| 130 | offsets = offsets.cumsum(dim=0) |
| 131 | |
| 132 | files_embed = self.file_embedding_bag(flat_indexes, offsets) |
| 133 | files_embed = self.dense_files(files_embed) |
| 134 | files_embed = self.activation_fn(files_embed) |
| 135 | files_embed = self.dropout(files_embed) |
| 136 | files_embed = self.out_proj_files(files_embed) |
| 137 | |
| 138 | # Add author embedding |
| 139 | authors: List[str] = input_batch.author |
| 140 | author_ids = [ |
| 141 | self.author_map.get(author, self.author_map[UNKNOWN_TOKEN]) |
| 142 | for author in authors |
| 143 | ] |
| 144 | author_ids = torch.tensor(author_ids).to(device) |
| 145 | author_embed = self.embedding_table(author_ids) |
| 146 | author_embed = self.dense_author(author_embed) |
| 147 | author_embed = self.activation_fn(author_embed) |
| 148 | author_embed = self.dropout(author_embed) |
| 149 | author_embed = self.out_proj_author(author_embed) |
| 150 | |
| 151 | return title_embed + files_embed + author_embed |
| 152 | |
| 153 | def convert_index_to_category_name(self, most_likely_index): |
no test coverage detected