Why do fit_one_cycle( ) repeat a similar result in per 5 loops

Part 1 (2019)
1

hello,when i train a model in tabular ,i try to know how the network can go,so i write:

for i in range(10):
learn.fit_one_cycle(5, 2e-3,wd=0.2)
learn.recorder.plot_losses()
learn.save(‘L11111112_LOSS_CS_’+str(i))
learn.lr_find()
learn.recorder.plot()
print(‘L11111112_LOSS_CS_’+str(i))

and then i found that every 5 loops ,the output is similar …

i think it’s different from learn.fit_one_cycle(50, 2e-3,wd=0.2)

why ???why is the out put so similar in the for loops code ???

i catch the result in per 5 loops and a 20 loops result , we can find that : it’s different …

20_loops
20_loops.png990×791 41.6 KB

similar_5_loops_0
similar_5_loops_0.png961×931 40 KB

similar_5loops
similar_5loops.png945×1257 64.5 KB

2

when you save learn by writing learn.save(), you should write learn.load() to pick up what you have saved.

4

thanks for your answer!
so this say…
if i don’t write learn.load() after learn.save(),and i continue to run learn.fit_one_cycle(), the learn will again run from the beginning ?

5

I think so.

6

You may check this: https://docs.fast.ai/basic_train.html#Learner.save