Lesson 5 discussion

Part 1 (2017)
64

I tried cats and dogs redux using vgg16bn by

  • changing vgg=Vgg16() to vgg=Vgg16BN(),

  • Importing using the updated utils.py and vgg16bn.py

I have a few questions:

  1. My accuracies are about the same at 0.9755 for vgg16 and vgg16BN, is there anything I might have missed?

    fig 1
    fig 1.png862×500 29.1 KB

  2. At this cell, Is it fine to have output 0.999725 for for the first picture?

    fig 2
    fig 2.png850×393 164 KB

  3. Based on the following confusion matrices, is there anything that I might have missed when I ran vgg16BN? The confusion matrix after I used vgg16BN shows 1000 cats-cats and 947 dogs-dogs:

    fig 3
    fig 3.png766×640 41.9 KB

While the confusion matrix after I used vgg16 (without BN) shows 979 cats-cats and 986 dogs-dogs:

fig 4
fig 4.png762×598 37.8 KB

65

Are there any good rules about using multiple embeddings? For example, using word2vec or GloVe as a general embedding but also adding a domain-specific embedding for something like medical terminology.

66

Hi, bobbylindsey. I have the same problem .Did you figure it out?

67

I got it . Thx!!! I thought vgg16() method create the file named vgg16.h5 , before. But I know I was wrong. ( I configure the running environment on my own computer ,not AWS. So, I didn’t have the model files before)

68

Is it pre-training model ?