The recall result of Memorize is higher than TPPRL in reproduction.

[L-M-Sherlock]

I try to update the code from tensorflow 1.8 to 2.15: https://github.com/L-M-Sherlock/tpprl

I reproduce the spaced repetition experimental and get this following result:

The result in paper is:

The recall of Memorize in my reproduction is higher than the result in paper.

My scripts:

python train-teaching.py ./data/initial_difficulty.csv 0.049 0.0052 ./output-spaced-repetition --q 0.0001 --q-entropy 0.005

python analyze-spaced-repetition.py ./data/initial_difficulty.csv 0.049 0.0052 ./output-spaced-repetition/save

The output of analyze-spaced-repetition.py in shell is:

WARNING:tensorflow:From /Users/jarrettye/Codes/tpprl_v2/tpprl/exp_teacher.py:470: dynamic_rnn (from tensorflow.python.ops.rnn) is deprecated and will be removed in a future version.
Instructions for updating:
Please use `keras.layers.RNN(cell)`, which is equivalent to this API
2022-12-21 14:03:50.928775: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:357] MLIR V1 optimization pass is not enabled
2022-12-21 14:03:51.228100: W tensorflow/tsl/platform/profile_utils/cpu_utils.cc:128] Failed to get CPU frequency: 0 Hz
[2022-12-21T14:03:51] Restored successfully to step 20050.
*** 2022-12-21 14:04:00.268328 Checkpoint: Initial capacity = 385.4, target capacity = 1875.46, q_init = 1.0
*** 2022-12-21 14:04:00.425130 Checkpoint: q = 2.0, capacity = 601.6
*** 2022-12-21 14:04:00.712795 Checkpoint: q = 4.0, capacity = 873.8
*** 2022-12-21 14:04:01.198034 Checkpoint: q = 8.0, capacity = 1199.3
*** 2022-12-21 14:04:02.087590 Checkpoint: q = 16.0, capacity = 1573.25
*** 2022-12-21 14:04:03.871065 Checkpoint: q = 32.0, capacity = 1988.5
*** 2022-12-21 14:04:03.871098 Checkpoint: q_hi = 16.0, q_lo = 32.0
*** 2022-12-21 14:04:05.388291 Checkpoint: new_capacity = 1813.85, q = 24.0
*** 2022-12-21 14:04:07.163421 Checkpoint: new_capacity = 1905.05, q = 28.0
*** 2022-12-21 14:04:08.694893 Checkpoint: new_capacity = 1860.35, q = 26.0
[2022-12-21T14:04:18] Done.

Could you help me check this problem?

[musically-ut]

Thanks for the translation @L-M-Sherlock!

Were you able to reproduce the results on the older version of Tensorflow, before the translation?

I'm travelling at the moment, let me get to a stable keyboard before I try to dig into the code.

[L-M-Sherlock]

Thanks for your reply. I make the translation because the older version of TensorFlow does not support my m1 mac. I will try to find another supported device to reproduce the result.

Have a good trip!