Question

我从以下来电中收到以下消息：

import pyspark.ml.regression as sparkreg
...
aftsr = sparkreg.AFTSurvivalRegression(labelCol="time", censorCol="event", featuresCol="features", maxIter=1000)
aftsr = aftsr.fit(sdf)

2018-04-17 14:20:42 ERROR StrongWolfeLineSearch:27 - Encountered bad values in function evaluation. Decreasing step size to 0.5
2018-04-17 14:20:43 ERROR StrongWolfeLineSearch:27 - Encountered bad values in function evaluation. Decreasing step size to 0.25
2018-04-17 14:20:43 ERROR LBFGS:27 - Failure! Resetting history: breeze.optimize.FirstOrderException: Line search zoom failed
2018-04-17 14:20:43 ERROR LBFGS:27 - Failure! Resetting history: breeze.optimize.FirstOrderException: Line search failed
2018-04-17 14:20:43 ERROR LBFGS:27 - Failure again! Giving up and returning. Maybe the objective is just poorly behaved?

然后它返回缩放和拦截的“坏”值。有办法避免这种情况吗？我可以向.fit提供初始估算值吗？不幸的是，我无法分享数据。这在一个问题实例中发生，但在其他问题实例中不会发生。

Answer 1

由Breeze库中的另一个错误（SPARK-21523）引起的已知错误（SPARK-21919，SPARK-21919，Fix strong wolfe line search init value bug #651）。

应该在Spark 2.2.1和2.3中修复。

AFTSurvivalRegression LBFGS错误目标表现不佳

1 个答案: