dest_evaluate can't work with dest.bin

[qianyunw]

I tried different max depths with "dest_train --load-mirrored --load-max-size 640 --rectangles helen\trainset100\rectangles.csv --create-num-shapes 60 --train-num-pixels 600 --train-num-splits 40 --train-max-depth X --train-learn 0.15 helen\trainset100"

but when train-max-depth > 6, dest_evaluate can't work, and it said "Failed to load tracker."

It has bothered me for three days, thank you so much!

[cheind]

If it can't load the tracker it seems to be that the tracker hasn't been written to disk. Could you verify if training gives you any error or does not write the tracking file?

[qianyunw]

Thank you so much! I found it is because vertifier._max_tables in dest_evaluate is smaller than that in dest.bin. When I set _max_tables to a bigger number, It works! Thank you!

[cheind]

hmm, interesting. what exactly did you do? Could you share a code snippet?

wangdayan notifications@github.com schrieb am Mo., 12. März 2018 um 06:40 Uhr:

Thank you so much! I found it is because vertifier._max_tables in dest_evaluate is smaller than that in dest.bin. When I set _max_tables to a bigger number, It works! Thank you!

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/cheind/dest/issues/15#issuecomment-372199073, or mute the thread https://github.com/notifications/unsubscribe-auth/AACWpaKX4WO0jdZkF2AO5Tzw_THzRHqrks5tdgpZgaJpZM4SlQWm .

[qianyunw]

Sure. In tracker.cpp I changed from: "flatbuffers::Verifier v(reinterpret_cast<const uint8_t>(buf.data()), buf.size());" to "flatbuffers::Verifier v(reinterpret_cast<const uint8_t>(buf.data()), buf.size(), 64, 900000000000);"

[cheind]

Thanks, I will consider changing the code, or, at least keep this in the back of my mind.