Pytorch-geomretic学习笔记 params_count,sparseTensor

最新推荐文章于 2024-08-06 20:28:49 发布

LUQC638

最新推荐文章于 2024-08-06 20:28:49 发布

阅读量1.8k

点赞数 2

分类专栏： pytorch-geometric graph neural network 文章标签： pytorch-geome pytorch 深度学习

本文链接：https://blog.youkuaiyun.com/weixin_61445075/article/details/124116495

版权

params_count(model),model是nn.Module类型的，计算参数的数量。

SparseTensor:

rom torch_sparse import SparseTensor

adj = SparseTensor(row=edge_index[0], col=edge_index[1], value=...,
                   sparse_sizes=(num_nodes, num_nodes))
# value is optional and can be None

# Obtain different representations (COO, CSR, CSC):
row,    col, value = adj.coo()
rowptr, col, value = adj.csr()
colptr, row, value = adj.csc()

adj = adj[:100, :100]  # Slicing, indexing and masking support
adj = adj.set_diag()   # Add diagonal entries
adj_t = adj.t()        # Transpose
out = adj.matmul(x)    # Sparse-dense matrix multiplication
adj = adj.matmul(adj)  # Sparse-sparse matrix multiplication

# Creating SparseTensor instances:
adj = SparseTensor.from_dense(mat)
adj = SparseTensor.eye(100, 100)
adj = SparseTensor.from_scipy(mat)

支持sparse-sparse,sparse-dense矩阵乘法。

在消息传递的时候