File size: 4,380 Bytes
9f0f10c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 | """
Dataloader for global_data_v2.npz.
Returns all fields needed by mjm_2a/b/c trainers.
"""
import numpy as np
import torch
from torch.utils.data import Dataset, DataLoader
class SpatialSEAAD_V2_Dataset(Dataset):
def __init__(self, X, y_class, y_subclass, y_supertype, batch_donor,
spatial, spatial_tiled, cps, confidence,
depth_norm, has_depth,
cell_volume_norm, has_volume,
meta_info):
self.X = torch.from_numpy(X).float()
self.y_class = torch.from_numpy(y_class).long()
self.y_subclass = torch.from_numpy(y_subclass).long()
self.y_supertype = torch.from_numpy(y_supertype).long()
self.batch_donor = torch.from_numpy(batch_donor).long()
self.spatial = torch.from_numpy(spatial).float()
self.spatial_tiled = torch.from_numpy(spatial_tiled).float()
self.cps = torch.from_numpy(cps).float()
self.confidence = torch.from_numpy(confidence).float()
self.depth_norm = torch.from_numpy(depth_norm).float()
self.has_depth = torch.from_numpy(has_depth.astype(np.float32)).float()
self.cell_volume_norm = torch.from_numpy(cell_volume_norm).float()
self.has_volume = torch.from_numpy(has_volume.astype(np.float32)).float()
meta = meta_info.item()
self.num_c = meta.get('num_class')
self.num_sc = meta.get('num_subclass')
self.num_st = meta.get('num_supertype')
def __len__(self):
return self.X.shape[0]
def __getitem__(self, idx):
return {
'X': self.X[idx],
'spatial': self.spatial[idx],
'spatial_tiled': self.spatial_tiled[idx],
'batch_id': self.batch_donor[idx],
'y_class': self.y_class[idx],
'y_subclass': self.y_subclass[idx],
'y_supertype': self.y_supertype[idx],
'cps': self.cps[idx],
'confidence': self.confidence[idx],
'depth_norm': self.depth_norm[idx],
'has_depth': self.has_depth[idx],
'cell_volume_norm': self.cell_volume_norm[idx],
'has_volume': self.has_volume[idx],
}
def build_dataloaders_v2(npz_path, batch_size=512, seed=42, num_workers=8):
data = np.load(npz_path, allow_pickle=True)
X_all = data['X']
y_c_all = data['y_class']
y_sc_all = data['y_subclass']
y_st_all = data['y_supertype']
batch_all = data['batch_donor']
spatial_all = data['spatial']
spatial_tiled_all = data['spatial_tiled']
cps_all = data['cps']
conf_all = data['y_supertype_confidence']
depth_all = data['depth_norm']
has_depth_all = data['has_depth']
vol_all = data['cell_volume_norm']
has_vol_all = data['has_volume']
meta_info = data['meta']
# per-donor split (same logic as v1)
unique_donors = np.unique(batch_all)
np.random.seed(seed)
np.random.shuffle(unique_donors)
test_donors = unique_donors[:6]
val_donors = unique_donors[6:9]
train_donors = unique_donors[9:]
def _slice(mask):
return SpatialSEAAD_V2_Dataset(
X_all[mask], y_c_all[mask], y_sc_all[mask], y_st_all[mask],
batch_all[mask], spatial_all[mask], spatial_tiled_all[mask],
cps_all[mask], conf_all[mask],
depth_all[mask], has_depth_all[mask],
vol_all[mask], has_vol_all[mask],
meta_info,
)
mask_train = np.isin(batch_all, train_donors)
mask_val = np.isin(batch_all, val_donors)
mask_test = np.isin(batch_all, test_donors)
train_ds = _slice(mask_train)
val_ds = _slice(mask_val)
test_ds = _slice(mask_test)
train_loader = DataLoader(train_ds, batch_size=batch_size, shuffle=True,
drop_last=True, num_workers=num_workers)
val_loader = DataLoader(val_ds, batch_size=batch_size, shuffle=True,
num_workers=num_workers)
test_loader = DataLoader(test_ds, batch_size=batch_size, shuffle=False,
num_workers=num_workers)
return train_loader, val_loader, test_loader
|