Makemore is a tool/program that generates more of what we give. Here, a names dataset (consisting of 32k names) was used with intention to produce more unique names. To start with, a bigram model is being put to use. A bigram model is a language model that learns all pairs of occurrences of letters to predict the next letter. Key takeaways:
- Extraction of Bigrams from the given dataset
- Visualisation of the bigrams
- Generation of text using a bigram model
- Broadcasting of the tensors (Need to practise more of this, can lead to a lot of bugs if not implemented properly)
- Understanding how loss functions can be used to evaluate the working of the bigram model
- Introduction to neural networks