Update app.py

app.py

@@ -4,9 +4,76 @@ import tensorflow as tf
 
 model = tf.saved_model.load('arabert_pretrained')
 
-def sentiment_analysis(text):
-    prediction = model.predict(text)
-    return prediction
+
+
+import pandas as pd
+df = pd.read_csv('put\data_cleaned1.csv')
+
+
+from transformers import TFAutoModel, AutoTokenizer
+arabert_tokenizer = AutoTokenizer.from_pretrained('aubmindlab/bert-base-arabert')
+
+
+
+import pandas as pd
+
+# Assuming your DataFrame is named 'df'
+# Split the DataFrame into two parts: label=1 and label=0
+label_1_df = df[df['data_labels'] == 1]
+label_0_df = df[df['data_labels'] == 0]
+
+# Sample an equal number of rows from each label
+sample_size = min(len(label_1_df), len(label_0_df))
+sample_label_1 = label_1_df.sample(n=sample_size, random_state=42)
+sample_label_0 = label_0_df.sample(n=sample_size, random_state=42)
+
+# Concatenate the two samples to get the final balanced sample
+balanced_sample = pd.concat([sample_label_1, sample_label_0])
+
+# Shuffle the rows in the balanced sample
+balanced_sample = balanced_sample.sample(frac=1, random_state=42)
+
+
+balanced_sample.reset_index(inplace=True,drop=True)
+
+
+from sklearn.model_selection import train_test_split
+
+tweets = balanced_sample['cleaned_text']
+labels = balanced_sample['data_labels']
+
+X_train, X_test, y_train, y_test = train_test_split(tweets, labels,stratify=labels, test_size=0.15, random_state=1)
+def preprocess_input_data(texts, tokenizer, max_len=120):
+    """Tokenize and preprocess the input data for Arabert model.
+
+    Args:
+        texts (list): List of text strings.
+        tokenizer (AutoTokenizer): Arabert tokenizer from transformers library.
+        max_len (int, optional): Maximum sequence length. Defaults to 120.
+
+    Returns:
+        Tuple of numpy arrays: Input token IDs and attention masks.
+    """
+    # Tokenize the text data using the tokenizer
+    tokenized_data = [tokenizer.encode_plus(
+        t,
+        max_length=max_len,
+        pad_to_max_length=True,
+        add_special_tokens=True) for t in texts]
+
+    # Extract tokenized input IDs and attention masks
+    input_ids = [data['input_ids'] for data in tokenized_data]
+    attention_mask = [data['attention_mask'] for data in tokenized_data]
+
+    return input_ids, attention_mask
+def sentiment_analysis(text): 
+    X_input_ids, X_attention_mask = preprocess_input_data(text, arabert_tokenizer)
+    predictions = modelez(X_input_ids)
+    a=predictions.numpy()
+    return a
+
+import gradio as gr
+
 
 iface = gr.Interface(fn=sentiment_analysis, inputs="text", outputs="text")
 iface.launch()