search

MicrobeLab / DeepMicrobes

DeepMicrobes: taxonomic classification for metagenomics with deep learning
https://doi.org/10.1093/nargab/lqaa009
Apache License 2.0
81 stars 21 forks source link

Question about the size of the model #29

Open viccc17 opened 8 months ago

viccc17 commented 8 months ago

Hi! May I know the size of the DeepMicrobes model (the best performing one)? Like how many trainable parameters are there? Thank you!

MicrobeLab commented 8 months ago

Hi, the number of trainable parameters is mainly dependent on the vocabulary size and the embedding layer harbors the highest number of parameters. For example, the 12-mer model will have ~8M*100 parameters in the embedding layer and the number of parameters in other layers is minor.

viccc17 commented 8 months ago

Thanks for your reply! Actually, I'm trying to show the total parameters of the model for kmer=12 using model.summary(). However, it seems that the model is not defined using keras. I also cannot get the total parameter using tf.global_variables(). Can you help me with this? Thank you!

MicrobeLab commented 8 months ago

Maybe you can try: total_params = tf.reduce_sum([tf.reduce_prod(var.shape) for var in tf.trainable_variables()])