Orthogonal Model of Emotions

A Text Classifier created using TensorFlow and Keras

Author

C.J. Pitchford

Published

18 August 2025

Model and Weights

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 embedding (Embedding)       (None, 1000, 64)          6400000   
                                                             
 bidirectional (Bidirection  (None, 1000, 128)         66048     
 al)                                                             
                                                             
 global_max_pooling1d (Glob  (None, 128)               0         
 alMaxPooling1D)                                                 
                                                             
 dense (Dense)               (None, 64)                8256      
                                                             
 dropout (Dropout)           (None, 64)                0         
                                                             
 dense_1 (Dense)             (None, 47)                3055      
                                                             
=================================================================
Total params: 6477359 (24.71 MB)
Trainable params: 6477359 (24.71 MB)
Non-trainable params: 0 (0.00 Byte)

Usage

import numpy as np
import tensorflow as tf

import tensorflow.keras.preprocessing.text as text

import pickle
from tensorflow.keras.preprocessing.sequence import pad_sequences

# 1. Load pre-trained model
model = tf.keras.models.load_model('OME4tf/ome-4a-model.h5')

# 2. Load tokenizer and label encoder
with open('OME4tf/ome-4a-tokenizer.pkl', 'rb') as f:
    tokenizer = pickle.load(f)
with open('OME4tf/ome-4a-label_encoder.pkl', 'rb') as f:
    label_encoder = pickle.load(f)

# 3. Test model with prediction on text "I failed to hide my distress."
text = "I failed to hide my distress."
text_seq = tokenizer.texts_to_sequences([text])
max_len = 1000
text_seq = pad_sequences(text_seq, maxlen=max_len, padding='post')
pred_probs = model.predict(text_seq)
pred_label = np.argmax(pred_probs, axis=1)
print(f"Statement: {text}\nPrediction: {label_encoder.classes_[pred_label][0]}")

Additional

Tokenizer and label encoder included as JSON to avoid using pickle files.