release-pypi-0.1.12

README.md

@@ -86,6 +86,11 @@ second contains the corresponding query homologous genes.
 > for code testing. The resulting annotation accuracy may not be as good as 
 > using the full dataset as the reference.
 
+**Suggestions**
+> 
+> If you have sufficient GPU memory, setting the hidden-size `h_dim=512` 
+in "came/PARAMETERS.py" may result in a more accurate cell-type transfer. 
+
 ### Test CAME's pipeline (optional)
 
 To test the package, run the python file `test_pipeline.py`:
@@ -120,7 +125,7 @@ If you are having issues, please let us know. We have a mailing list located at:
 
 ### Citation
 
-If CAME is useful for your research, consider citing our preprint:
+If CAME is useful for your research, consider citing our work:
 
 > Liu X, Shen Q, Zhang S. Cross-species cell-type assignment of single-cell RNA-seq by a heterogeneous graph neural network[J]. Genome Research, 2022: gr. 276868.122.
 

README_CH.md

@@ -73,6 +73,10 @@ python setup.py install
 > 数据文件 “raw-Baron_human.h5ad” 仅用于代码测试，是原始数据的子样本 （20%），
 > 因此结果的注释精度可能不如使用完整数据集作为参考。
 
+### **建议**
+
+如果你有足够的GPU显存，我们建议在``came/PARAMETERS.py``中设置`h_dim=512` 来获得更好的结果。
+
 ### 测试 CAME 的分析流程 (非必要)
 
 可以直接运行 `test_pipeline.py` 来测试 CAME 的分析流程：

came/__init__.py

@@ -49,4 +49,4 @@
 from .pipeline import KET_CLUSTER, __test1__, __test2__
 
 
-__version__ = "0.1.10"
+__version__ = "0.1.12"

came/model/_utils.py

@@ -21,12 +21,17 @@
 from .cgc import CGCNet
 # from ._minibatch import create_blocks, create_batch
 
+try:
+    from dgl.dataloading import NodeDataLoader
+except ImportError:
+    from dgl.dataloading import DataLoader as NodeDataLoader
+
 
 def idx_hetero(feat_dict, id_dict):
     sub_feat_dict = {}
     for k, ids in id_dict.items():
         if k in feat_dict:
-            sub_feat_dict[k] = feat_dict[k][ids]
+            sub_feat_dict[k] = feat_dict[k][ids.cpu()]
         else:
             # logging.warning(f'key "{k}" does not exist in {feat_dict.keys()}')
             pass
@@ -158,7 +163,7 @@ def get_all_hidden_states(
             sampler = dgl.dataloading.MultiLayerNeighborSampler(
                 sampler.fanouts[:-1])
 
-        dataloader = dgl.dataloading.NodeDataLoader(
+        dataloader = NodeDataLoader(
             g, {'cell': g.nodes('cell'), 'gene': g.nodes('gene')},
             sampler, device=device,
             batch_size=batch_size, shuffle=False, drop_last=False, num_workers=0
@@ -301,7 +306,7 @@ def get_model_outputs(
         ######################################
         if sampler is None:
             sampler = model.get_sampler(g.canonical_etypes, 50)
-        dataloader = dgl.dataloading.NodeDataLoader(
+        dataloader = NodeDataLoader(
             g, {'cell': g.nodes('cell')},
             sampler, device=device,
             batch_size=batch_size,

came/utils/train.py

@@ -32,6 +32,11 @@
 from .plot import plot_records_for_trainer
 from ._base_trainer import BaseTrainer, SUBDIR_MODEL
 
+try:
+    from dgl.dataloading import NodeDataLoader
+except ImportError:
+    from dgl.dataloading import DataLoader as NodeDataLoader
+
 
 def seed_everything(seed=123):
     """ not works well """
@@ -52,7 +57,7 @@ def make_class_weights(labels, astensor=True, foo=np.sqrt, n_add=0):
 
     counts = value_counts(labels).sort_index()  # sort for alignment
     n_cls = len(counts) + n_add
-    w = counts.apply(lambda x: 1 / foo(x + 1) if x > 0 else 0)  
+    w = counts.apply(lambda x: 1 / foo(x + 1) if x > 0 else 0)
     w = (w / w.sum() * (1 - n_add / n_cls)).values
     w = np.array(list(w) + [1 / n_cls] * int(n_add))
 
@@ -117,7 +122,7 @@ def prepare4train(
         test_idx = LongTensor(test_idx)
 
     g = dpair.get_whole_net(rebuild=False, )
-    g.nodes[node_cls_type].data[key_label] = labels   # date: 211113
+    g.nodes[node_cls_type].data[key_label] = labels  # date: 211113
 
     ENV_VARs = dict(
         classes=classes,
@@ -134,8 +139,8 @@ def prepare4train(
 
 class Trainer(BaseTrainer):
     """
-    
-    
+
+
     """
 
     def __init__(self,
@@ -309,7 +314,7 @@ def train(self, n_epochs=350,
                 **params_lossfunc
             )
 
-            # prediction 
+            # prediction
             _, y_pred = torch.max(logits, dim=1)
 
             # ========== evaluation (Acc.) ==========
@@ -367,22 +372,24 @@ def train_minibatch(self, n_epochs=100,
         if sampler is None:
             sampler = model.get_sampler(g.canonical_etypes, 50)
 
-        train_dataloader = dgl.dataloading.NodeDataLoader(
+        train_dataloader = NodeDataLoader(
             # The following arguments are specific to NodeDataLoader.
-            g, {'cell': train_idx},  # The node IDs to iterate over in minibatches
-            sampler, device=device,  # Put the sampled MFGs on CPU or GPU
+            g, {'cell': train_idx},
+            # The node IDs to iterate over in minibatches
+            sampler, device='cpu',  # Put the sampled MFGs on CPU or GPU
             # The following arguments are inherited from PyTorch DataLoader.
             batch_size=batch_size, shuffle=True, drop_last=False, num_workers=0
         )
-        test_dataloader = dgl.dataloading.NodeDataLoader(
-            g, {'cell': test_idx}, sampler, device=device, batch_size=batch_size,
+        test_dataloader = NodeDataLoader(
+            g, {'cell': test_idx}, sampler, device='cpu', batch_size=batch_size,
             shuffle=False, drop_last=False, num_workers=0
         )
         print(f" start training (device='{device}') ".center(60, '='))
+        rcd = {}
         for epoch in range(n_epochs):
             model.train()
             self._cur_epoch += 1
-            
+
             t0 = time.time()
             all_train_preds = []
             train_labels = []
@@ -443,6 +450,7 @@ def train_minibatch(self, n_epochs=100,
                 **rcd, print_info=self._cur_epoch % info_stride == 0 or backup)
         self.log_info(**rcd, print_info=True)
         self._cur_epoch_adopted = self._cur_epoch
+        self.save_checkpoint_record()
 
     def get_current_outputs(self,
                             feat_dict=None,
@@ -526,7 +534,8 @@ def evaluate_metrics(
         )
         return metrics
 
-    def log_info(self, train_acc, test_acc, ami=None, print_info=True, # dur=0.,
+    def log_info(self, train_acc, test_acc, ami=None, print_info=True,
+                 # dur=0.,
                  **kwargs):
         dur_avg = np.average(self.dur)
         ami = kwargs.get('AMI', 'NaN') if ami is None else ami
@@ -575,7 +584,7 @@ def infer_for_nodes(
     if reorder:
         return order_by_ids(all_test_preds, orig_ids)
     return all_test_preds
-        
+
 
 def order_by_ids(x, ids):
     """reorder by the original ids"""

setup.py

@@ -20,7 +20,7 @@
 EMAIL = '[email protected]'
 AUTHOR = 'Xingyan Liu'
 REQUIRES_PYTHON = '>=3.8.0'
-VERSION = '0.1.10'
+VERSION = '0.1.12'
 
 REQUIRED = [
     'scanpy',