Datasets
Models
Diff of
/datasets/abc_dataset.py
[000000]
..
[03464c]
Switch to unified view
| a | b/datasets/abc_dataset.py | ||
|---|---|---|---|
| 1 | import os.path |
||
| 2 | from datasets import load_file |
||
| 3 | from datasets import get_survival_y_true |
||
| 4 | from datasets.basic_dataset import BasicDataset |
||
| 5 | from util import preprocess |
||
| 6 | import numpy as np |
||
| 7 | import pandas as pd |
||
| 8 | import torch |
||
| 9 | |||
| 10 | |||
| 11 | class ABCDataset(BasicDataset): |
||
| 12 | """ |
||
| 13 | A dataset class for multi-omics dataset. |
||
| 14 | For gene expression data, file should be prepared as '/path/to/data/A.tsv'. |
||
| 15 | For DNA methylation data, file should be prepared as '/path/to/data/B.tsv'. |
||
| 16 | For miRNA expression data, file should be prepared as '/path/to/data/C.tsv'. |
||
| 17 | For each omics file, each columns should be each sample and each row should be each molecular feature. |
||
| 18 | """ |
||
| 19 | |||
| 20 | def __init__(self, param): |
||
| 21 | """ |
||
| 22 | Initialize this dataset class. |
||
| 23 | """ |
||
| 24 | BasicDataset.__init__(self, param) |
||
| 25 | self.omics_dims = [] |
||
| 26 | |||
| 27 | # Load data for A |
||
| 28 | A_df = load_file(param, 'A') |
||
| 29 | # Get the sample list |
||
| 30 | if param.use_sample_list: |
||
| 31 | sample_list_path = os.path.join(param.data_root, 'sample_list.tsv') # get the path of sample list |
||
| 32 | self.sample_list = np.loadtxt(sample_list_path, delimiter='\t', dtype='<U32') |
||
| 33 | else: |
||
| 34 | self.sample_list = A_df.columns |
||
| 35 | # Get the feature list for A |
||
| 36 | if param.use_feature_lists: |
||
| 37 | feature_list_A_path = os.path.join(param.data_root, 'feature_list_A.tsv') # get the path of feature list |
||
| 38 | feature_list_A = np.loadtxt(feature_list_A_path, delimiter='\t', dtype='<U32') |
||
| 39 | else: |
||
| 40 | feature_list_A = A_df.index |
||
| 41 | A_df = A_df.loc[feature_list_A, self.sample_list] |
||
| 42 | self.A_dim = A_df.shape[0] |
||
| 43 | self.sample_num = A_df.shape[1] |
||
| 44 | A_array = A_df.values |
||
| 45 | if self.param.add_channel: |
||
| 46 | # Add one dimension for the channel |
||
| 47 | A_array = A_array[np.newaxis, :, :] |
||
| 48 | self.A_tensor_all = torch.Tensor(A_array) |
||
| 49 | self.omics_dims.append(self.A_dim) |
||
| 50 | |||
| 51 | # Load data for B |
||
| 52 | B_df = load_file(param, 'B') |
||
| 53 | # Get the feature list for B |
||
| 54 | if param.use_feature_lists: |
||
| 55 | feature_list_B_path = os.path.join(param.data_root, 'feature_list_B.tsv') # get the path of feature list |
||
| 56 | feature_list_B = np.loadtxt(feature_list_B_path, delimiter='\t', dtype='<U32') |
||
| 57 | else: |
||
| 58 | feature_list_B = B_df.index |
||
| 59 | B_df = B_df.loc[feature_list_B, self.sample_list] |
||
| 60 | if param.ch_separate: |
||
| 61 | B_df_list, self.B_dim = preprocess.separate_B(B_df) |
||
| 62 | self.B_tensor_all = [] |
||
| 63 | for i in range(0, 23): |
||
| 64 | B_array = B_df_list[i].values |
||
| 65 | if self.param.add_channel: |
||
| 66 | # Add one dimension for the channel |
||
| 67 | B_array = B_array[np.newaxis, :, :] |
||
| 68 | B_tensor_part = torch.Tensor(B_array) |
||
| 69 | self.B_tensor_all.append(B_tensor_part) |
||
| 70 | else: |
||
| 71 | self.B_dim = B_df.shape[0] |
||
| 72 | B_array = B_df.values |
||
| 73 | if self.param.add_channel: |
||
| 74 | # Add one dimension for the channel |
||
| 75 | B_array = B_array[np.newaxis, :, :] |
||
| 76 | self.B_tensor_all = torch.Tensor(B_array) |
||
| 77 | self.omics_dims.append(self.B_dim) |
||
| 78 | |||
| 79 | # Load data for C |
||
| 80 | C_df = load_file(param, 'C') |
||
| 81 | # Get the feature list for C |
||
| 82 | if param.use_feature_lists: |
||
| 83 | feature_list_C_path = os.path.join(param.data_root, 'feature_list_C.tsv') # get the path of feature list |
||
| 84 | feature_list_C = np.loadtxt(feature_list_C_path, delimiter='\t', dtype='<U32') |
||
| 85 | else: |
||
| 86 | feature_list_C = C_df.index |
||
| 87 | C_df = C_df.loc[feature_list_C, self.sample_list] |
||
| 88 | self.C_dim = C_df.shape[0] |
||
| 89 | C_array = C_df.values |
||
| 90 | if self.param.add_channel: |
||
| 91 | # Add one dimension for the channel |
||
| 92 | C_array = C_array[np.newaxis, :, :] |
||
| 93 | self.C_tensor_all = torch.Tensor(C_array) |
||
| 94 | self.omics_dims.append(self.C_dim) |
||
| 95 | |||
| 96 | self.class_num = 0 |
||
| 97 | if param.downstream_task == 'classification': |
||
| 98 | # Load labels |
||
| 99 | labels_path = os.path.join(param.data_root, 'labels.tsv') # get the path of the label |
||
| 100 | labels_df = pd.read_csv(labels_path, sep='\t', header=0, index_col=0).loc[self.sample_list, :] |
||
| 101 | self.labels_array = labels_df.iloc[:, -1].values |
||
| 102 | # Get the class number |
||
| 103 | self.class_num = len(labels_df.iloc[:, -1].unique()) |
||
| 104 | elif param.downstream_task == 'regression': |
||
| 105 | # Load target values |
||
| 106 | values_path = os.path.join(param.data_root, 'values.tsv') # get the path of the target value |
||
| 107 | values_df = pd.read_csv(values_path, sep='\t', header=0, index_col=0).loc[self.sample_list, :] |
||
| 108 | self.values_array = values_df.iloc[:, -1].astype(float).values |
||
| 109 | self.values_max = self.values_array.max() |
||
| 110 | self.values_min = self.values_array.min() |
||
| 111 | elif param.downstream_task == 'survival': |
||
| 112 | # Load survival data |
||
| 113 | survival_path = os.path.join(param.data_root, 'survival.tsv') # get the path of the survival data |
||
| 114 | survival_df = pd.read_csv(survival_path, sep='\t', header=0, index_col=0).loc[self.sample_list, :] |
||
| 115 | self.survival_T_array = survival_df.iloc[:, -2].astype(float).values |
||
| 116 | self.survival_E_array = survival_df.iloc[:, -1].values |
||
| 117 | self.survival_T_max = self.survival_T_array.max() |
||
| 118 | self.survival_T_min = self.survival_T_array.min() |
||
| 119 | if param.survival_loss == 'MTLR': |
||
| 120 | self.y_true_tensor = get_survival_y_true(param, self.survival_T_array, self.survival_E_array) |
||
| 121 | if param.stratify_label: |
||
| 122 | labels_path = os.path.join(param.data_root, 'labels.tsv') # get the path of the label |
||
| 123 | labels_df = pd.read_csv(labels_path, sep='\t', header=0, index_col=0).loc[self.sample_list, :] |
||
| 124 | self.labels_array = labels_df.iloc[:, -1].values |
||
| 125 | elif param.downstream_task == 'multitask': |
||
| 126 | # Load labels |
||
| 127 | labels_path = os.path.join(param.data_root, 'labels.tsv') # get the path of the label |
||
| 128 | labels_df = pd.read_csv(labels_path, sep='\t', header=0, index_col=0).loc[self.sample_list, :] |
||
| 129 | self.labels_array = labels_df.iloc[:, -1].values |
||
| 130 | # Get the class number |
||
| 131 | self.class_num = len(labels_df.iloc[:, -1].unique()) |
||
| 132 | |||
| 133 | # Load target values |
||
| 134 | values_path = os.path.join(param.data_root, 'values.tsv') # get the path of the target value |
||
| 135 | values_df = pd.read_csv(values_path, sep='\t', header=0, index_col=0).loc[self.sample_list, :] |
||
| 136 | self.values_array = values_df.iloc[:, -1].astype(float).values |
||
| 137 | self.values_max = self.values_array.max() |
||
| 138 | self.values_min = self.values_array.min() |
||
| 139 | |||
| 140 | # Load survival data |
||
| 141 | survival_path = os.path.join(param.data_root, 'survival.tsv') # get the path of the survival data |
||
| 142 | survival_df = pd.read_csv(survival_path, sep='\t', header=0, index_col=0).loc[self.sample_list, :] |
||
| 143 | self.survival_T_array = survival_df.iloc[:, -2].astype(float).values |
||
| 144 | self.survival_E_array = survival_df.iloc[:, -1].values |
||
| 145 | self.survival_T_max = self.survival_T_array.max() |
||
| 146 | self.survival_T_min = self.survival_T_array.min() |
||
| 147 | if param.survival_loss == 'MTLR': |
||
| 148 | self.y_true_tensor = get_survival_y_true(param, self.survival_T_array, self.survival_E_array) |
||
| 149 | elif param.downstream_task == 'alltask': |
||
| 150 | # Load labels |
||
| 151 | self.labels_array = [] |
||
| 152 | self.class_num = [] |
||
| 153 | for i in range(param.task_num-2): |
||
| 154 | labels_path = os.path.join(param.data_root, 'labels_'+str(i+1)+'.tsv') # get the path of the label |
||
| 155 | labels_df = pd.read_csv(labels_path, sep='\t', header=0, index_col=0).loc[self.sample_list, :] |
||
| 156 | self.labels_array.append(labels_df.iloc[:, -1].values) |
||
| 157 | # Get the class number |
||
| 158 | self.class_num.append(len(labels_df.iloc[:, -1].unique())) |
||
| 159 | |||
| 160 | # Load target values |
||
| 161 | values_path = os.path.join(param.data_root, 'values.tsv') # get the path of the target value |
||
| 162 | values_df = pd.read_csv(values_path, sep='\t', header=0, index_col=0).loc[self.sample_list, :] |
||
| 163 | self.values_array = values_df.iloc[:, -1].astype(float).values |
||
| 164 | self.values_max = self.values_array.max() |
||
| 165 | self.values_min = self.values_array.min() |
||
| 166 | |||
| 167 | # Load survival data |
||
| 168 | survival_path = os.path.join(param.data_root, 'survival.tsv') # get the path of the survival data |
||
| 169 | survival_df = pd.read_csv(survival_path, sep='\t', header=0, index_col=0).loc[self.sample_list, :] |
||
| 170 | self.survival_T_array = survival_df.iloc[:, -2].astype(float).values |
||
| 171 | self.survival_E_array = survival_df.iloc[:, -1].values |
||
| 172 | self.survival_T_max = self.survival_T_array.max() |
||
| 173 | self.survival_T_min = self.survival_T_array.min() |
||
| 174 | if param.survival_loss == 'MTLR': |
||
| 175 | self.y_true_tensor = get_survival_y_true(param, self.survival_T_array, self.survival_E_array) |
||
| 176 | |||
| 177 | def __getitem__(self, index): |
||
| 178 | """ |
||
| 179 | Return a data point and its metadata information. |
||
| 180 | |||
| 181 | Returns a dictionary that contains A_tensor, B_tensor, C_tensor, label and index |
||
| 182 | input_omics (list) -- a list of input omics tensor |
||
| 183 | label (int) -- label of the sample |
||
| 184 | index (int) -- the index of this data point |
||
| 185 | """ |
||
| 186 | # Get the tensor of A |
||
| 187 | if self.param.add_channel: |
||
| 188 | A_tensor = self.A_tensor_all[:, :, index] |
||
| 189 | else: |
||
| 190 | A_tensor = self.A_tensor_all[:, index] |
||
| 191 | |||
| 192 | # Get the tensor of B |
||
| 193 | if self.param.ch_separate: |
||
| 194 | B_tensor = [] |
||
| 195 | for i in range(0, 23): |
||
| 196 | if self.param.add_channel: |
||
| 197 | B_tensor_part = self.B_tensor_all[i][:, :, index] |
||
| 198 | else: |
||
| 199 | B_tensor_part = self.B_tensor_all[i][:, index] |
||
| 200 | B_tensor.append(B_tensor_part) |
||
| 201 | # Return a list of tensor |
||
| 202 | else: |
||
| 203 | if self.param.add_channel: |
||
| 204 | B_tensor = self.B_tensor_all[:, :, index] |
||
| 205 | else: |
||
| 206 | B_tensor = self.B_tensor_all[:, index] |
||
| 207 | # Return a tensor |
||
| 208 | |||
| 209 | # Get the tensor of C |
||
| 210 | if self.param.add_channel: |
||
| 211 | C_tensor = self.C_tensor_all[:, :, index] |
||
| 212 | else: |
||
| 213 | C_tensor = self.C_tensor_all[:, index] |
||
| 214 | |||
| 215 | if self.param.downstream_task == 'classification': |
||
| 216 | # Get label |
||
| 217 | label = self.labels_array[index] |
||
| 218 | return {'input_omics': [A_tensor, B_tensor, C_tensor], 'label': label, 'index': index} |
||
| 219 | elif self.param.downstream_task == 'regression': |
||
| 220 | # Get target value |
||
| 221 | value = self.values_array[index] |
||
| 222 | return {'input_omics': [A_tensor, B_tensor, C_tensor], 'value': value, 'index': index} |
||
| 223 | elif self.param.downstream_task == 'survival': |
||
| 224 | # Get survival T and E |
||
| 225 | survival_T = self.survival_T_array[index] |
||
| 226 | survival_E = self.survival_E_array[index] |
||
| 227 | y_true = self.y_true_tensor[index, :] |
||
| 228 | return {'input_omics': [A_tensor, B_tensor, C_tensor], 'survival_T': survival_T, 'survival_E': survival_E, 'y_true': y_true, 'index': index} |
||
| 229 | elif self.param.downstream_task == 'multitask': |
||
| 230 | # Get label |
||
| 231 | label = self.labels_array[index] |
||
| 232 | # Get target value |
||
| 233 | value = self.values_array[index] |
||
| 234 | # Get survival T and E |
||
| 235 | survival_T = self.survival_T_array[index] |
||
| 236 | survival_E = self.survival_E_array[index] |
||
| 237 | y_true = self.y_true_tensor[index, :] |
||
| 238 | return {'input_omics': [A_tensor, B_tensor, C_tensor], 'label': label, 'value': value, |
||
| 239 | 'survival_T': survival_T, 'survival_E': survival_E, 'y_true': y_true, 'index': index} |
||
| 240 | elif self.param.downstream_task == 'alltask': |
||
| 241 | # Get label |
||
| 242 | label = [] |
||
| 243 | for i in range(self.param.task_num - 2): |
||
| 244 | label.append(self.labels_array[i][index]) |
||
| 245 | # Get target value |
||
| 246 | value = self.values_array[index] |
||
| 247 | # Get survival T and E |
||
| 248 | survival_T = self.survival_T_array[index] |
||
| 249 | survival_E = self.survival_E_array[index] |
||
| 250 | y_true = self.y_true_tensor[index, :] |
||
| 251 | return {'input_omics': [A_tensor, B_tensor, C_tensor], 'label': label, 'value': value, |
||
| 252 | 'survival_T': survival_T, 'survival_E': survival_E, 'y_true': y_true, 'index': index} |
||
| 253 | else: |
||
| 254 | return {'input_omics': [A_tensor, B_tensor, C_tensor], 'index': index} |
||
| 255 | |||
| 256 | def __len__(self): |
||
| 257 | """ |
||
| 258 | Return the number of data points in the dataset. |
||
| 259 | """ |
||
| 260 | return self.sample_num |
||
| 261 |