Add pre-computed models that have already been computed so students could load the models for machines without cuda

For 4_Chinese_document_classification, I had to add an escape on the '%' when defining the chars variable in "2. Build vocab" to use it when implementing a solution using the names.

Some of the rows have no name, but a language.

In names_train.csv: line 983, 2148, 2991, 3248, 4494, 4904, 5972, 7260, 7955, 8122, 9779, 10408, 10696

In names_test.csv: line 983, 2148

To add a leading . in the path like

from .datautils.vocabulary import Vocabulary

to avoid conflict with the one in the public repo.

also, datautils.misc is missing in
day_2/02_Names_CharNN_Nationality_Classifier.ipynb

• Needed to add "import json" to perform save in TweetLanguageModel and TrumpTweetVectorizer
• Needed to call "model.cuda()" to perform training
• " ".join(greedy_sample_from(model, vectorizer, temperature=0.9, n_length=30, use_cuda=False)[1:]) needs to be moved below creation of "model" object.

• in NamesClassifier class: "last_item_indices += torch.arange(0, x_in.size(0)).long().cuda() * x_in.size(1)" needs a non-gpu version (and should be pushed into the "if use_gpu" check.
• need "import json" for save