POEM GENERATOR GENERATOR

I could tweak the hyperparameters of my LSTM model, e.g.

I could also use additional features of the poems to improve the quality of the model’s output, e.g.

I could use additional seeds to improve relevance e.g.

Currently, the model uses a starting word as the seed for generating the rest of the poem.

In addition to the observations above, there are several other possible directions for this project in the future:

Both models are limited by the vocabulary of the dataset a possible next step could be to extend this vocabulary to include what the Stanford GloVe model contains.

100 dimensions) to see how the results are affected. Given the 300-fold increase in number of features due to the inclusion of the 300-dimension word embeddings, this is unsurprising.

The GloVe approach resulted in a model that was more than 1 GB in size, which compares to the cleaned tokens approach that was a ‘mere’ 81 MB in size.

That said, as the video illustrates, even the second approach wasn’t foolproof, and on average it didn’t seem to produce ‘decent’ poetry more than 50% of the time.

The second approach using GloVe embeddings appears to produce natural-sounding poems more often than the first approach using cleaned tokens.

POEM GENERATOR GENERATOR

A temperature of 0 would mean that the generator chooses the most ‘likely’ word every time, while higher temperatures gives the generator an increasing likelihood of choosing a less ‘likely’ word. The ‘creative temperature’ correlates to how ‘adventerous’ the poetry generator is when choosing the next word of the poem. Next steps would be to try additional seeds, additional features, different model hyperparameters, and a larger and/or improved poetry corpus.Training on cleaned word sequences produced poetry that didn’t quite make sense training with GloVe embeddings produced much better results, but inconsistently.Trained an LSTM model with two approaches: cleaned word sequences and raw word sequences paired with Stanford’s GloVe embeddings.This time round, my aim is to generate short poetry by feeding a poetry corpus into a Long-Short-Term Memory (LSTM) neural network.

Identifying fake news: The LIAR dataset and its limitations.Is a trawler fishing? Modelling the Global Fishing Watch dataset.MyAnimeList user scores: Fun with web scraping and linear regression.MTA Turnstile Data: My First Taste of a Data Science Project.The first four projects I’ve done are as follows: And here we are, the final project of my Metis Data Science Bootcamp.

Poem generator