-
Notifications
You must be signed in to change notification settings - Fork 2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
XGBoost dist=gamma reports NA performance metrics #7832
Comments
Megan Kurka commented: Possibly related: [https://github.com/dmlc/xgboost/issues/5349|https://github.com/dmlc/xgboost/issues/5349|smart-link] ? |
Megan Kurka commented: Note: H2O GBM with Gamma converges for me |
Jan Sterba commented: reproduced with native xgboost, updated issue linked above with our finding, not much more we can do |
Neema Mashayekhi commented: Reopening till native XGB is confirmed that this is the case. May be able to find workaround from dmlc |
Neema Mashayekhi commented: [~accountid:5c2ff866e7af4d69171f81d9] - have you seen this before? Any ways to prevent it? |
Jan Sterba commented: fixed upstream, will add to our builds [https://github.com/dmlc/xgboost/pull/6537|https://github.com/dmlc/xgboost/pull/6537|smart-link] |
Jan Sterba commented: tried to verify the upstream fix, but it did not work reported a new issue here: [https://github.com/dmlc/xgboost/issues/6723|https://github.com/dmlc/xgboost/issues/6723|smart-link] |
JIRA Issue Migration Info Jira Issue: PUBDEV-7810 Attachments From Jira Attachment Name: pyunit_gamma_zero_metrics.py |
For the following example, the deviance grows with each tree until validation metrics report NA's. I was able to reproduce this on a single or multi-node cluster on 3.26.0.2.
{code:python}
df = h2o.import_file("hdfs://mr-0xd6.0xdata.loc:8020/datasets/airlines_all.05p.csv")
df["Year"] = (df["Year"] < 2000).ifelse(0, df["Year"])
train, valid = df.split_frame(seed=1234)
from h2o.estimators.xgboost import H2OXGBoostEstimator
xgb = H2OXGBoostEstimator(model_id="xgb1.hex",
col_sample_rate=0.7,
stopping_rounds=5,
stopping_metric="deviance",
stopping_tolerance=0.00001,
distribution="gamma",
ntrees=1000,
score_tree_interval=5,
)
xgb.train(training_frame=train, validation_frame=valid, y="Year")
{code}
The text was updated successfully, but these errors were encountered: