581. Google - Fast or Slow? Predict AI Model Runtime | predict-ai-model-runtime
感谢主办方的比赛邀请,祝贺所有获奖者!
我将简要分享我的解决方案。
def forward(self, batch):
node_opcode = batch.node_opcode.long()
opcode_embeds = self.opcode_embedding(node_opcode)
x = torch.concat([batch.node_feat, opcode_embeds, batch.node_config_feat * self.node_config_weights], dim=1)
x = self.lin1(x)
x = self.norm1(x).relu()
x_init = x
for i in range(self.n_layers):
x = self.convs[i](x, batch.edge_index)
x = self.norms[i](x).relu()
x = x_init + x
x = torch.concat([global_mean_pool(x, batch.batch), global_max_pool(x, batch.batch)], dim=1)
x = self.dropout(x)
x = self.readout(x)
return x
针对不同子类型分别训练模型
交叉验证分数(使用官方提供的训练/验证分割)
基于这些交叉验证分数,我最终获得了公开榜0.684分和私有榜0.688分的成绩。