Skip to content

Kiana-Jahanshid/Emoji-Text-Classification

Repository files navigation

Emoji Classification : 😄🧡🍴⚾😞

Description :

Here , we have used glove.6B.zip pre-trained word vectors To sum the feature vector corresponding to each word in each sentence .
We summed the feature vectors of all of the words in every sentence . As a result, we have a single summed feature vector for 'each sentence' of the test and train files.
Finally, we feed the summed feature vectors corresponding to each sentence , to the network , as new form of X_train and X_test .

How to install

pip -r install requirements.txt

How to run :

you should write your desired sentence and dimension like below command :

python emoji_classification.py --sentence "I Love AI" --dimension 200  

Results :

without Dropout :

Feature vectors Dimension Train Loss Train Accuracy Test Loss Test Accuracy Inference time
50d 0.47 0.89 0.56 0.87 0.0460 second
100d 0.33 0.94 0.5 0.83 0.0468 second
200d 0.16 0.98 0.43 0.80 0.0474 second
300d 0.09 1.0 0.39 0.82 0.0489 second

With Dropout = 0.4 :

Feature vectors Dimension Train Loss Train Accuracy Test Loss Test Accuracy Inference time
50d 0.79 0.70 0.71 0.78 0.0458 second
100d 0.64 0.78 0.63 0.82 0.0476 second
200d 0.36 0.89 0.45 0.83 0.049 second
300d 0.20 0.95 0.44 0.85 0.0498 second
  • Dropout conclusion :

  • 🔺 Using Dropout causes higher values in "Test Accuracy" in higher dimensions;and resulting in better performance on unseen data(in larger dimensions)

  • 🔻 But instead with Dropout , "Train accuracy" values have been decreased in all dimensions .

  • ⚠ Using Dropout , increases both train and test loss values .

Releases

No releases published

Packages

No packages published