Context This dataset contains spaCy embeddings and their respective target labels for 1.6m tweets corresponding to an existing raw dataset that was published on Kaggle by Μαριο? Μιχαηλιδη? KazAnova around three years ago. From my personal experience with Kaggle kernels, spaCy seems to work quite slow on notebooks relative to my personal machine. So I believe this dataset can become really handy if you are training your own machine learning and deep learning models as it simply eliminates the first step of the process i.e. pre-processing 1.6m tweets and converting them to tweet embeddings ready to be used for learning without any hassle! Content The original dataset in its raw form can be accessed using this [link](https://www.kaggle.com/kazanova/sentiment140). Here, the dataset is divided into two files. The first `Spacy_training_full.pickle` has a shape of 1.6m x 300 as it contains standardized vectors for each tweet converted using spaCy `en_vectors_web_lg` the largest industrial strength language library provided by spaCy. A python file showing a variation for the same has been hosted on GitHub and can be accessed using this [link](https://github.com/akshaydnicator/Twitter-Sentiment-Analysis-Competition---Analytics-Vidhya/blob/master/Data_Preprocessing/Tweets_to_Spacy_embeddings.py). The second file `Spacy_targets_training_full.pickle` has corresponding target labels for the training. It is has been pre-processed using keras.utils to_categorical and can be used directly with keras. For the sake of simplicity, the order of the tweet embeddings is not changed and is kept as it is in the orginal raw dataset i.e. First 800,000 rows are labeled as '0' showing a negative sentiment and the last 800,000 rows are labeled as '1' showing positive sentiment. Acknowledgements Thanks to [Μαριο? Μιχαηλιδη? KazAnova](https://www.kaggle.com/kazanova) for the original raw tweet dataset. Inspiration Any new ideas are welcomed. For starters, it would be great to know how useful spaCy embeddings can be in machine learning and deep learning models on kaggle and what is the best accuracy that can be achieved using the huge lableled ready to use tweet embeddings public dataset.