Error raised when predicting or dumping count:poisson model

[yangs16]

terminate called after throwing an instance of 'dmlc::ParamError'
what(): Required parameter max_delta_step of float is not presented
Aborted

---

An easy fix is to add a config item after the line 265 in src/learner.cc, e.g.
cfg_["max_delta_step"] = common::ToString(0.7);

Thanks.

[khotilov]

xgboost/src/learner.cc

Line 168 in 49bbd72

cfg_["max_delta_step"] = "0.7";

should have taken care of that.
Could you please show how to reproduce your error?

[yangs16]

Yes. But the Configure() is not called in CLIDump2Text() and CLIPredict(). I guess the fix should be add a line after

xgboost/src/cli_main.cc

Line 271 in 49bbd72

std::unique_ptr<Learner> learner(Learner::Create({}));

and

xgboost/src/cli_main.cc

Line 297 in 49bbd72

std::unique_ptr<Learner> learner(Learner::Create({}));

:

learner->Configure(param.cfg);

Hope this will help.

[yangs16]

Just noticed that the problem with the prediction has been fixed in commit 1dde863. I send another pull request to fix the problem with the model dump: #1253