1. add one more import:
   from torch.nn import Parameter

2. At In[16] and In[17], in case you don't have the preprocessed pkl files. (guess you won't)

def loadGloveModel(gloveFile):
print "Loading Glove Model"
f = open(gloveFile,'r')
model = {}
for line in f:
splitLine = line.split()
word = splitLine[0]
embedding = np.array([float(val) for val in splitLine[1:]], dtype='f')
model[word] = embedding
print "Done.",len(model)," words loaded!"
return model

gloveFile='./datasets/nlp/glove.6B.100d.txt'
en_w2v = loadGloveModel(gloveFile)
dim_en_vec = 100
n_en_vec = len(en_w2v)

Hi，
def load_glove(loc):
return (bcolz.open(loc+'.dat')[:],
pickle.load(open(loc+'_words.pkl','rb'), encoding='latin1'),
pickle.load(open(loc+'_idx.pkl','rb'), encoding='latin1'))
en_vecs, en_wv_word, en_wv_idx = load_glove('/data/datasets/nlp/glove/results/6B.100d')
en_w2v = {w: en_vecs[en_wv_idx[w]] for w in en_wv_word}
n_en_vec, dim_en_vec = en_vecs.shape
I can't find 6B.100d_words.pkl, 6B.100d_idx.pkl where https://nlp.stanford.edu/projects/glove/. Can you tell download them from where, thanks.

I am modifying bits of the code so that it works with keras 2. Would you be interested in PRs?

Hi, Jeremy,

In the lesson 1, the vgg16.py has a extra Lambda layer for the vgg_preprocess which minus the rgb mean and revert the RGB to BGR. Do we still need the color RGB preprocess with the new vgg16 in this repo?

Thanks
tjliupeng

Hey, what's up?
Thanks so much for uploading the code.

regarding this line:
def Var(*sz): return Parameter(Arr(*sz)).cuda()
I believe cudaing this decoder parameters is redundant as we are cudaing the decoder later on in the code.