example about run linear on yarn about difacto_dmlc HOT 7 CLOSED

@formath src/linear/run_yarn.sh is not ok?

from difacto_dmlc.

@CNevd I can't find "guide/demo_hdfs.conf" in run_yarn.sh. Is this configure file the hdfs-site.conf?

from difacto_dmlc.

@formath guide/demo_hdfs.conf is generated by
cat <<< " train_data = \"hdfs://${hdfs_dir}/data/agaricus.txt.train\" val_data = \"hdfs://${hdfs_dir}/data/agaricus.txt.test\" max_data_pass = 3 " >guide/demo_hdfs.conf
or you can write your own .conf

from difacto_dmlc.

@CNevd Yeah. It's simple to use. But I still want to know could linear in this project support millions features and instances? If based on ps-lite, I think it will be.

from difacto_dmlc.

yes, it supports, we use linear and difacto for millions features and 10~100 millions instances

from difacto_dmlc.

@CNevd Great job. The original project of difacto does't support distributed training. Is this project support that? What is the main difference against the original project?

from difacto_dmlc.

it supports distributed training， this project is extract from wormhole/learn/difacto which is the first version of difacto

from difacto_dmlc.