Closed Martinn1996 closed 1 year ago
Hi @Martinn1996 ,
I have added an example here. You can set ''mode'' to ''vertical'' and ''partition_mode'' to "vertical" to simulate the vertical FL.
Thank you! I think there is an error then in the documentation, since it says that the possible value for partition_mode are iid
or noniid
I was also wondering how I can make the partition myself? As in, can I decide which features belongs to which party, thanks in advance.
Hi @Martinn1996 ,
Thanks for the information! I have corrected the document. For the partition, you can manually partition the dataset into multiple subsets and use the CLI interface to conduct training by specifying them as data
. For example, the configuration file example.conf
is
data=./dataset/credit/credit_vertical_p0_withlabel.csv,./dataset/credit/credit_vertical_p1.csv
test_data=./dataset/credit/credit_vertical_p0_withlabel.csv,./dataset/credit/credit_vertical_p1.csv
model_path=fedtree.model
n_class=2
data_format=csv
n_parties=2
objective=binary:logistic
mode=vertical
privacy_tech=none
n_trees=10
depth=6
max_num_bin=16
learning_rate=0.1
Then run ./build/bin/FedTree-train example.conf
.
We'll integrate this feature into Python interface. Thanks!
Hi, can you provide an example using the Python wrapper for Vertical FL? Currently the examples all make use of Horizontal FL.