A tool to give feedback on your writing in real time. It is focused on providing parents a less subjective way to evaluate writing while making it fun for students to imitate authors they like.

Current Status: In developement. Sentences are passed to a tensorflow model that outputs a sentence embedding. See project board for how you can contribute.

Live Site

To host this for free and give live feedback on the writing this tool is built as as a static page using tensorflow.js where are all calculations are performed on the client.

• app/ - Single page app hosted on github (https://wroscoe.github.io/gradest/) that runs text through a tensorflow.js model and returns a grade.
• collector/ - Scripts to aggregate the text and text attributes (writer age, publication, writer name, ect) used to help train the tensorflow model.
• trainer/ - Scripts to train the tensorflow model and convert it to tensorflow.js.
• Makefile - Place for frequently used commands.

• Old books that are out of print.
• Blog posts
• Writing samples by students.

• Writing Grade Level/Age - Use the age of the author as the target variable.
• Most similar author. - What author is most similar to the users writing.
• Writing period - What decade is the users writing like.

• Most word embedding models are too big (>100MB) to load in a static webpage. The lite word embedding I found only works on TF1. How can we create our own lite word/sentence embedding that works in TF2 or keras?

• What is the right way to train the network so that it can output many different sparse metrics. Some datasets may not have the age of the author.

• Keras Text Classification Models
• Keras to tensorlfow.js
• convert keras model to tensorflow.js
• tensorflow universal sentence embeder with colab example, requires TF1
• tensorflow_js using word embeddings