search

geffy / tffm

TensorFlow implementation of an arbitrary order Factorization Machine
MIT License
780 stars 176 forks source link

Does tffm support multi-threading? #13

Closed binga closed 7 years ago

binga commented 7 years ago

Does tffm support multi-threading? While running tffm, all the cores on my machine are not running at 100% CPU. Is there anyway to add multi-threading at tf-level into tffm just like how it's configured on keras here: https://github.com/fchollet/keras/blob/master/keras/backend/tensorflow_backend.py#L105 ? Thank you.

geffy commented 7 years ago

Hi @binga! I have only 2 cores on my laptop and it seems that TF use all of them at 100%. Maybe it depends on blas/atlas compilation options? But indeed, you can pass intra_op_parallelism_threads param into TFFM constructor in a similar way as in gpu_benchmark.ipynb:

nb_thread = 4
config = tf.ConfigProto(intra_op_parallelism_threads=nb_thread, allow_soft_placement=True)
model = TFFMClassifier(
    order=3,
    rank=10, 
    optimizer=tf.train.AdamOptimizer(learning_rate=0.01), 
    n_epochs=100, 
    batch_size=-1,
    init_std=0.001,
    reg=0.001,
    input_type='dense',
    session_config=config,
    log_dir='./tmp'
)
geffy commented 7 years ago

Please re-open if the problem is still not resolved.