如何在 Python 中在 H2O 上启用 GPU 加速？

Question

我正在尝试使用 Python 和 H2O 在我的 AWS 服务器上训练深度学习神经网络，并且我想启用 GPU 加速以加快训练速度。请让我知道使用 GPU 而不是 CPU 的代码片段。AWS 使用 OpenGL。弹性 GPU 类型是具有 4 GB 内存的 eg1.xlarge。

我的模型的代码是：

nn = H2OGridSearch(model=H2ODeepLearningEstimator,
                               hyper_params = {
        'activation' :[ "Rectifier","Tanh","Maxout","RectifierWithDropout","TanhWithDropout","MaxoutWithDropout"],
        'hidden':[[20,20],[50,50],[30,30,30],[25,25,25,25]],            ## small network, runs faster
#        'rate' :[0.0005,0.001,0.0015,0.002,0.0025,0.003,0.0035,0.0040,0.0045,0.005],
         'l1':[0,1e-4,1e-6],
         'l2':[0,1e-4,1e-6]
        })

start_time = time.time()

nn.train(
        train1_x, train1_y,train1,
        score_validation_samples = 10000,      ## sample the validation dataset (faster)
        stopping_rounds = 2,
        stopping_metric ="MSE", ## alternatives: "MSE","logloss","r2"
        epochs=1000000,  
        stopping_tolerance = 0.01,
        max_w2 = 10
        )
end_time = time.time()

Answer 1

如果您有兴趣运行支持 GPU 的 H2O.ai 产品，H2O-3 仅启用 CPU，请参阅H2O4GPU或 Driverless AI（注意：后者是闭源的）