R中H2O深度学习网格搜索遇CURL超时错误求助

问题场景 #

在R 4.2.0版本环境中运行H2O深度学习二分类模型的网格搜索，使用的代码如下：

hyper_params <- list(
    activation = c("Rectifier", "Maxout", "Tanh", "RectifierWithDropout", "MaxoutWithDropout", "TanhWithDropout"),
    hidden = list(c(5, 5, 5, 5, 5), c(10, 10, 10, 10), c(50, 50, 50), c(100, 100, 100)),
    epochs = c(50, 100, 200),
    l1 = c(0, 0.00001, 0.0001),
    l2 = c(0, 0.00001, 0.0001),
    rate = c(0, 01, 0.005, 0.001),
    rate_annealing = c(1e-8, 1e-7, 1e-6),
    rho = c(0.9, 0.95, 0.99, 0.999),
    epsilon = c(1e-10, 1e-8, 1e-6, 1e-4),
    momentum_start = c(0, 0.5),
    momentum_stable = c(0.99, 0.5, 0),
    input_dropout_ratio = c(0, 0.1, 0.2),
    max_w2 = c(10, 100, 1000, 3.4028235e+38)
)
search_criteria <- list(strategy = "RandomDiscrete",
                                                max_models = 100,
                                                max_runtime_secs = 12000,
                                                stopping_tolerance = 0.001,
                                                stopping_rounds = 15,
                                                seed = 42)
dl_grid <- h2o.grid(algorithm = "deeplearning",
                                        x = x,
                                        y = y,
                                        grid_id = "dl_grid",
                                        training_frame = df,
                                        validation_frame = tst,
                                        nfolds = 25,                          
                                        fold_assignment = "Stratified",
                                        hyper_params = hyper_params,
                                        search_criteria = search_criteria,
                                        seed = 42
)

运行后出现以下错误：

"na .h2o.doSafeREST(h2oRestApiVersion = h2oRestApiVersion, urlSuffix = urlSuffix, : Unexpected CURL error: Timeout was reached: [localhost:54321] Resolving timed out after 10180 milliseconds [1] "Job request failed Unexpected CURL error: Timeout was reached: [localhost:54321] Resolving timed out after 10180 milliseconds, will retry after 3s."

使用的R版本详情：

platform       x86_64-w64-mingw32
arch           x86_64
os             mingw32
crt            ucrt
system         x86_64, mingw32
status
major          4
minor          2.0
year           2022
month          04
day            22
svn rev        82229
language       R
version.string R version 4.2.0 (2022-04-22 ucrt)
nickname       Vigorous Calisthenics

解决方法 #

• 修正参数语法错误：代码中rate参数里的01是无效值，应改为0.01，避免因参数错误导致模型训练异常。
• 延长H2O超时时间：初始化H2O时增加超时参数，比如h2o.init(timeout = 30000)（单位为毫秒），给API调用足够的等待时间。
• 简化超参数空间：当前超参数组合数量过大，即使随机选100个模型，计算量也极高。建议先只保留关键参数（如activation、hidden、epochs）的候选值，其他参数用默认值，逐步缩小搜索范围。
• 降低交叉验证折数：25折交叉验证会大幅增加每个模型的训练时间，建议先降到5或10折，待找到较优参数范围后再提升折数。
• 增加H2O内存分配：重启H2O时指定更大内存，比如h2o.init(max_mem_size = "8G")（根据机器实际内存调整），避免因内存不足导致训练卡顿超时。
• 检查本地端口与防火墙：确保localhost的54321端口未被防火墙或安全软件拦截，保证R与H2O服务的正常通信。
• 重启H2O服务：先用h2o.shutdown()关闭现有服务，再用h2o.init()重新启动，清除可能的资源占用问题。

内容的提问来源于stack exchange，提问作者Achu A L