So I have been training my tensorflow model on a list of 400.000 tweets, of which half is positive, the other half negative. All went well: the accuracy is around 0.77.
W
