I tried to train a model using the settings mentioned in lesson 1. I was able to go through till the end, but have a few questions to check if i was right.

When i plotted the lr curve i got something like this.

now i tried to to do this with varying learning rates

case 1 : learn.fit_one_cycle(2, max_lr=slice(1e-3,1e-2))
Got bad results, high error rate (5.something)

Case 2: learn.fit_one_cycle(2, max_lr=slice(1e-4,1e-2))
still bad results high error rate (0.14)

case 3 : learn.fit_one_cycle(2, max_lr=slice(1e-6,1e-3))
got 0.0000 error rate ? - > is this normal ? Does it mean i messed something up or in the given data its perform 100% ? Or does it just mean my validation dataset is too little (it has 5 images per category) and my training set has 30 per category.

if my model is indeed pretty good, how can i now deploy it to test out new images ? Is there a platform where i can deploy my model and ask people for feedback ?

Hi, @tarun98601. I recommend you to continue studying deep learning. Lesson 2 gives answers about using trained model in production.

Your dataset seems to be very small and the model might not give the good results on practice. Depends on the data though.
I started with 81 images for classifying 4 categories and could not train the model at all. Then it started training with 320 images. I also had error_rate = 0 at the end. But the model still gives not perfect results. I continued studying lessons and there was a way to draw heatmap. I saw which features of the image were used and why the model was often wrong.

@kushaj i did reset the weights, @tsar - thanks for your inputs, i will continue to lesson 2 then and see and keep this as “the” model i will work on and keep improving it.