加载GraphDTA 预训练模型进行预测出现的问题
收藏
走完一遍GraphDTA 模型训练并将best model保存为best_model.pdparams(只保存了state_dict)
# 保存模型代码如下:
paddle.save(model.state_dict(), best_model)
之后想要直接加载best_model.pdparams,用来预测药物与靶点蛋白之间的相互作用时却报错:
# 模型加载代码如下:
model = paddle.load("best_model.pdparams")
···
···
···
oin_graph = join_graph.tensor()
proteins_token = paddle.to_tensor(proteins_token)
proteins_mask = paddle.to_tensor(proteins_mask)
model.eval()
affinity_pred = model(join_graph, proteins_token, proteins_mask)
affinity_pred = affinity_pred.numpy()[0][0]
print('亲和力预测:', affinity_pred)
# 报错如下:
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
~6939.py in
43 proteins_mask = paddle.to_tensor(proteins_mask)
44
---> 45 model.eval()
46 affinity_pred = model(join_graph, proteins_token, proteins_mask)
47 affinity_pred = affinity_pred.numpy()[0][0]
AttributeError: 'dict' object has no attribute 'eval'
请问,是只有保存了完整的模型才可以直接加载用来预测相互作用嘛?
0
收藏
请登录后评论
问题已解决