test.py

from transformers import ConvBertTokenizer, TFConvBertModel
import tensorflow as tf
import numpy as np
from tensorflow.keras import backend as K
from tensorflow.keras import regularizers
import tensorflow as tf
from tensorflow.keras.layers import *
from tensorflow.keras.models import *
from transformers import *
import os
from text_cleaning import clean_text
from huggingface_hub import hf_hub_download


os.environ['CUDA_VISIBLE_DEVICES'] = '-1'

# gpu_number = 1 #### GPU number 
# gpus = tf.config.experimental.list_physical_devices('GPU')
# if gpus:
#     tf.config.experimental.set_visible_devices(gpus[gpu_number], 'GPU') 
#     logical_gpus = tf.config.experimental.list_logical_devices('GPU')
#     print(len(gpus), "Physical GPUs,", len(logical_gpus), "Logical GPU")

MAX_LENGTH = 32
BATCH_SIZE = 256

model_name = 'dbmdz/convbert-base-turkish-mc4-uncased'
tokenizer = ConvBertTokenizer.from_pretrained(model_name)
CUDA_VISIBLE_DEVICES=4

label_to_name = {0:"INSULT", 
                 1:"OTHER", 
                 2:"PROFANITY", 
                 3:"RACIST", 
                 4:"SEXIST"}


custom_object = {"TFConvBertModel": TFConvBertModel, "K":K}

second_model_1_path = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="2inci_model_mc4_emir_aug_data_dropout01_0.h5")
second_model_1 = tf.keras.models.load_model(second_model_1_path, custom_objects=custom_object, compile=False)

second_model_2_model_path = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="2inci_model_mc4_emir_aug_data_dropout01_1.h5")
second_model_2 = tf.keras.models.load_model(second_model_2_model_path, custom_objects=custom_object, compile=False)

second_model_3_model_path = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="2inci_model_mc4_emir_aug_data_dropout01_2.h5")
second_model_3 = tf.keras.models.load_model(second_model_3_model_path, custom_objects=custom_object, compile=False)

second_model_4_model_path = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="2inci_model_mc4_emir_aug_data_dropout01_3.h5")
second_model_4 = tf.keras.models.load_model(second_model_4_model_path, custom_objects=custom_object, compile=False)

second_model_5_model_path = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="2inci_model_mc4_emir_aug_data_dropout01_4.h5")
second_model_5 = tf.keras.models.load_model(second_model_5_model_path, custom_objects=custom_object, compile=False)

third_model_1_path = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="3uncu_model_mc4_emir_aug_data_0.h5")
third_model_1 = tf.keras.models.load_model(third_model_1_path, custom_objects=custom_object, compile=False)

third_model_2_path = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="3uncu_model_mc4_emir_aug_data_1.h5")
third_model_2 = tf.keras.models.load_model(third_model_2_path, custom_objects=custom_object, compile=False)

third_model_3_path = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="3uncu_model_mc4_emir_aug_data_2.h5")
third_model_3 = tf.keras.models.load_model(third_model_3_path, custom_objects=custom_object, compile=False)

third_model_4_path = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="3uncu_model_mc4_emir_aug_data_3.h5")
third_model_4 = tf.keras.models.load_model(third_model_4_path, custom_objects=custom_object, compile=False)

third_model_5_path = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="3uncu_model_mc4_emir_aug_data_4.h5")
third_model_5 = tf.keras.models.load_model(third_model_5_path, custom_objects=custom_object, compile=False)

model_path1 = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="model0.h5")
first_model_1 = tf.keras.models.load_model(model_path1, custom_objects=custom_object, compile=False)

model_path2 = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="model1.h5")
first_model_2 = tf.keras.models.load_model(model_path2, custom_objects=custom_object, compile=False)

model_path3 = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="model2.h5")
first_model_3 = tf.keras.models.load_model(model_path3, custom_objects=custom_object, compile=False)

model_path4 = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="model3.h5")
first_model_4 = tf.keras.models.load_model(model_path4, custom_objects=custom_object, compile=False)

model_path5 = hf_hub_download(repo_id="emirkocak/TRT_Data_Warriors_tackling_hate_speech", filename="model4.h5")
first_model_5 = tf.keras.models.load_model(model_path5, custom_objects=custom_object, compile=False)

def bert_encode(data):
    tokens = tokenizer.batch_encode_plus(data, max_length=MAX_LENGTH, padding='max_length', truncation=True)
    
    return tf.constant(tokens['input_ids'])


def test_predict(text):

    text = clean_text(text)
    
    test_encoded = bert_encode([text])
    test_dataset = (
        tf.data.Dataset
        .from_tensor_slices((test_encoded))
        .batch(BATCH_SIZE))

    y_kfold_second = 0
    y_kfold_third = 0
    y_kfold_first = 0

    for model in [second_model_1, second_model_2, second_model_3, second_model_4, second_model_5]:
        y_kfold_second += model.predict(test_dataset)

    for model in [third_model_1, third_model_2, third_model_3, third_model_4, third_model_5]:
        y_kfold_third += model.predict(test_dataset)

    for model in [first_model_1, first_model_2, first_model_3, first_model_4, first_model_5]:
        y_kfold_first += model.predict(test_dataset)

    y_pred_all = 0.39 * y_kfold_first / 5 + 0.38 * y_kfold_second / 5 + 0.23 * y_kfold_third / 5

    # y_pred_all = y_kfold_first
    preds = np.argmax(y_pred_all, 1)

    preds_names = [label_to_name[pred] for pred in preds]

    return preds_names