Update app.py

app.py

@@ -1,200 +1,200 @@
-import ast
-import os
-import pickle
-import random
-from datetime import datetime, timedelta
-
-import gradio as gr
-import pandas as pd
-from langchain_google_genai import ChatGoogleGenerativeAI
-from langchain_core.messages import HumanMessage, SystemMessage
-from pytrends.request import TrendReq
-
-from mlxtend.preprocessing import TransactionEncoder
-
-
-def convert_keywords_to_list(keywords_str):
-    try:
-        return ast.literal_eval(keywords_str)
-    except (SyntaxError, ValueError):
-        return []
-
-
-def convert_scores_to_list(scores_float):
-    try:
-        return ast.literal_eval(scores_float)
-    except (SyntaxError, ValueError):
-        return []
-
-video_df = pd.read_csv('video_df_complete.csv')
-video_df['keywords'] = video_df['keywords'].apply(convert_keywords_to_list)
-video_df['trend_scores'] = video_df['trend_scores'].apply(convert_scores_to_list)
-
-video_df['total_score'] = video_df['trend_scores'].apply(lambda x: sum(x) / len(x) if len(x) > 0 else 0)
-
-transactions = []
-for index, row in video_df.iterrows():
-    transactions.append(row['keywords'])
-
-te = TransactionEncoder()
-te_ary = te.fit(transactions).transform(transactions)
-df = pd.DataFrame(te_ary, columns=te.columns_)
-
-merged_df = pd.concat([df, video_df['total_score'], video_df['engagement_rate']], axis=1)
-
-rules = pd.read_csv('association_rules.csv')
-rules['antecedents'] = rules['antecedents'].apply(lambda x: list(eval(x)))
-rules['consequents'] = rules['consequents'].apply(lambda x: list(eval(x)))
-
-model_filename = os.path.join('regression_model_final.pkl')
-
-with open(model_filename, 'rb') as file:
-    model = pickle.load(file)
-
-llm = ChatGoogleGenerativeAI(model="gemini-1.5-pro", convert_system_message_to_human=True)
-
-def custom_predict(keywords, total_score):
-    """
-    Custom prediction function using the trained linear regression model.
-
-    Args:
-        keywords: A list of keywords.
-        total_score: The total trend score.
-
-    Returns:
-        The predicted engagement rate.
-    """
-    new_data = pd.DataFrame([{col: 0 for col in merged_df.columns}])
-
-    for keyword in keywords:
-        if keyword in new_data.columns:
-            new_data.at[0, keyword] = 1
-
-    new_data.at[0, 'total_score'] = total_score
-
-    new_data = new_data.drop('engagement_rate', axis=1)
-
-    prediction = model.predict(new_data)
-
-    return prediction[0][0]
-
-
-def generate_keyword_scores(keywords):
-    scaled_rate = min(100, 4.5 * 10)
-
-    return [
-        round(random.uniform(scaled_rate * 0.7, min(100, scaled_rate * 1.2)), 2)
-        for _ in keywords
-    ]
-
-
-def get_google_trends_score(keywords, end_date, days_back=7):
-    """
-    Mengambil skor tren Google untuk kata kunci tertentu selama periode waktu tertentu.
-
-    Parameters:
-    keywords (list): Daftar kata kunci yang ingin dianalisis.
-    end_date (datetime): Tanggal akhir untuk data tren.
-    days_back (int): Jumlah hari ke belakang dari end_date untuk menentukan rentang waktu (default: 7 hari).
-
-    Returns:
-    pd.DataFrame: DataFrame berisi data tren per kata kunci selama periode waktu yang ditentukan.
-    """
-    try:
-      if not keywords:
-          raise ValueError("Daftar kata kunci tidak boleh kosong.")
-
-      pytrends = TrendReq()
-      start_date = end_date - timedelta(days=days_back)
-      timeframe = f"{start_date.strftime('%Y-%m-%d')} {end_date.strftime('%Y-%m-%d')}"
-
-      pytrends.build_payload(keywords, timeframe=timeframe, geo='ID', gprop='youtube')
-      trends_df = pytrends.interest_over_time()
-
-      if 'isPartial' in trends_df.columns:
-          trends_df = trends_df.drop(columns=['isPartial'])
-
-      return trends_df
-    except Exception as e:
-      return pd.DataFrame(generate_keyword_scores(keywords))
-
-
-def generate_title(keyword, category):
-    if category != 'Gaming':
-        return "Category belum supported."
-
-    recommendation = recommend_keyword(keyword)
-
-    if not recommendation:
-        return "No recommendations found."
-    else:
-        result = llm(
-            [
-                SystemMessage(
-                    content=f"Kamu adalah seorang penulis judul video youtube"
-                            f"Kamu akan diberikan beberapa buah keyword yang wajib digunakan untuk judul"
-                            f"Buat judul yang semenarik mungkin untuk memberikan viewer rasa suka"
-                            f"Cukup keluarkan satu judul saja dalam satu kalimat"
-                            f"Jangan gunnakan formatting seperti '\n' atau hal lainnya. Gunakan saja raw string"
-                            f"Boleh pake emoji"
-                ),
-                HumanMessage(
-                    content=f"keyword yang digunakan adalah sebagai berikut: {recommendation}"
-                            f"Total jumlah keyword adalah: {len(recommendation)}"
-                            f"Video memiliki kategori: {category}"
-                )
-            ]
-        )
-
-        return result.content
-
-
-def recommend_keyword(keyword):
-    keyword_rules = rules[
-        rules['antecedents'].astype(str).str.contains(keyword) | rules['consequents'].astype(str).str.contains(keyword)]
-
-    top_5_rules = keyword_rules.sort_values(by='lift', ascending=False).head(5)
-
-    recommendation = []
-    engages = []
-
-    for idx, row in top_5_rules.iterrows():
-        antecedents = list(row['antecedents'])[0]
-        consequents = list(row['consequents'])
-
-        recommendation.append([keyword] + consequents)
-
-    if not recommendation:
-        return []
-
-    for rec in recommendation:
-        trends_df = get_google_trends_score(rec, datetime.now())
-
-        batch_scores = [
-            round(trends_df[keyword].mean(), 2) if keyword in trends_df.columns else 0
-            for keyword in keywords
-        ]
-
-        batch_scores = sum(batch_scores) / len(batch_scores)
-
-        engagement_rate = custom_predict(rec, batch_scores)
-
-        engages.append(engagement_rate)
-
-    return recommendation[engages.index(max(engages))]
-
-
-distinct_categories = video_df['catergory'].unique()
-
-iface = gr.Interface(
-    fn=generate_title,
-    inputs=[
-        gr.Textbox(label="Enter a keyword"),
-        gr.Dropdown(label="Select a category", choices=list(distinct_categories))
-    ],
-    outputs=gr.Textbox(label="Recommendations"),
-    title="Title Recommendation",
-    description="Do'akan saya langgeng sm Ei"
-)
-
-iface.launch()
+import ast
+import os
+import pickle
+import random
+from datetime import datetime, timedelta
+
+import gradio as gr
+import pandas as pd
+from langchain_google_genai import ChatGoogleGenerativeAI
+from langchain_core.messages import HumanMessage, SystemMessage
+from pytrends.request import TrendReq
+
+from mlxtend.preprocessing import TransactionEncoder
+
+
+def convert_keywords_to_list(keywords_str):
+    try:
+        return ast.literal_eval(keywords_str)
+    except (SyntaxError, ValueError):
+        return []
+
+
+def convert_scores_to_list(scores_float):
+    try:
+        return ast.literal_eval(scores_float)
+    except (SyntaxError, ValueError):
+        return []
+
+video_df = pd.read_csv('video_df_complete.csv')
+video_df['keywords'] = video_df['keywords'].apply(convert_keywords_to_list)
+video_df['trend_scores'] = video_df['trend_scores'].apply(convert_scores_to_list)
+
+video_df['total_score'] = video_df['trend_scores'].apply(lambda x: sum(x) / len(x) if len(x) > 0 else 0)
+
+transactions = []
+for index, row in video_df.iterrows():
+    transactions.append(row['keywords'])
+
+te = TransactionEncoder()
+te_ary = te.fit(transactions).transform(transactions)
+df = pd.DataFrame(te_ary, columns=te.columns_)
+
+merged_df = pd.concat([df, video_df['total_score'], video_df['engagement_rate']], axis=1)
+
+rules = pd.read_csv('association_rules.csv')
+rules['antecedents'] = rules['antecedents'].apply(lambda x: list(eval(x)))
+rules['consequents'] = rules['consequents'].apply(lambda x: list(eval(x)))
+
+model_filename = os.path.join('regression_model_final.pkl')
+
+with open(model_filename, 'rb') as file:
+    model = pickle.load(file)
+
+llm = ChatGoogleGenerativeAI(model="gemini-1.5-pro", convert_system_message_to_human=True)
+
+def custom_predict(keywords, total_score):
+    """
+    Custom prediction function using the trained linear regression model.
+
+    Args:
+        keywords: A list of keywords.
+        total_score: The total trend score.
+
+    Returns:
+        The predicted engagement rate.
+    """
+    new_data = pd.DataFrame([{col: 0 for col in merged_df.columns}])
+
+    for keyword in keywords:
+        if keyword in new_data.columns:
+            new_data.at[0, keyword] = 1
+
+    new_data.at[0, 'total_score'] = total_score
+
+    new_data = new_data.drop('engagement_rate', axis=1)
+
+    prediction = model.predict(new_data)
+
+    return prediction[0][0]
+
+
+def generate_keyword_scores(keywords):
+    scaled_rate = min(100, 4.5 * 10)
+
+    return [
+        round(random.uniform(scaled_rate * 0.7, min(100, scaled_rate * 1.2)), 2)
+        for _ in keywords
+    ]
+
+
+def get_google_trends_score(keywords, end_date, days_back=7):
+    """
+    Mengambil skor tren Google untuk kata kunci tertentu selama periode waktu tertentu.
+
+    Parameters:
+    keywords (list): Daftar kata kunci yang ingin dianalisis.
+    end_date (datetime): Tanggal akhir untuk data tren.
+    days_back (int): Jumlah hari ke belakang dari end_date untuk menentukan rentang waktu (default: 7 hari).
+
+    Returns:
+    pd.DataFrame: DataFrame berisi data tren per kata kunci selama periode waktu yang ditentukan.
+    """
+    try:
+      if not keywords:
+          raise ValueError("Daftar kata kunci tidak boleh kosong.")
+
+      pytrends = TrendReq()
+      start_date = end_date - timedelta(days=days_back)
+      timeframe = f"{start_date.strftime('%Y-%m-%d')} {end_date.strftime('%Y-%m-%d')}"
+
+      pytrends.build_payload(keywords, timeframe=timeframe, geo='ID', gprop='youtube')
+      trends_df = pytrends.interest_over_time()
+
+      if 'isPartial' in trends_df.columns:
+          trends_df = trends_df.drop(columns=['isPartial'])
+
+      return trends_df
+    except Exception as e:
+      return pd.DataFrame(generate_keyword_scores(keywords))
+
+
+def generate_title(keyword, category):
+    if category != 'Gaming':
+        return "Category belum supported."
+
+    recommendation = recommend_keyword(keyword)
+
+    if not recommendation:
+        return "No recommendations found."
+    else:
+        result = llm(
+            [
+                SystemMessage(
+                    content=f"Kamu adalah seorang penulis judul video youtube"
+                            f"Kamu akan diberikan beberapa buah keyword yang wajib digunakan untuk judul"
+                            f"Buat judul yang semenarik mungkin untuk memberikan viewer rasa suka"
+                            f"Cukup keluarkan satu judul saja dalam satu kalimat"
+                            f"Jangan gunnakan formatting seperti '\n' atau hal lainnya. Gunakan saja raw string"
+                            f"Boleh pake emoji"
+                ),
+                HumanMessage(
+                    content=f"keyword yang digunakan adalah sebagai berikut: {recommendation}"
+                            f"Total jumlah keyword adalah: {len(recommendation)}"
+                            f"Video memiliki kategori: {category}"
+                )
+            ]
+        )
+
+        return result.content
+
+
+def recommend_keyword(keyword):
+    keyword_rules = rules[
+        rules['antecedents'].astype(str).str.contains(keyword) | rules['consequents'].astype(str).str.contains(keyword)]
+
+    top_5_rules = keyword_rules.sort_values(by='lift', ascending=False).head(5)
+
+    recommendation = []
+    engages = []
+
+    for idx, row in top_5_rules.iterrows():
+        antecedents = list(row['antecedents'])[0]
+        consequents = list(row['consequents'])
+
+        recommendation.append([keyword] + consequents)
+
+    if not recommendation:
+        return []
+
+    for rec in recommendation:
+        trends_df = get_google_trends_score(rec, datetime.now())
+
+        batch_scores = [
+            round(trends_df[keyword].mean(), 2) if keyword in trends_df.columns else 0
+            for keyword in rec
+        ]
+
+        batch_scores = sum(batch_scores) / len(batch_scores)
+
+        engagement_rate = custom_predict(rec, batch_scores)
+
+        engages.append(engagement_rate)
+
+    return recommendation[engages.index(max(engages))]
+
+
+distinct_categories = video_df['catergory'].unique()
+
+iface = gr.Interface(
+    fn=generate_title,
+    inputs=[
+        gr.Textbox(label="Enter a keyword"),
+        gr.Dropdown(label="Select a category", choices=list(distinct_categories))
+    ],
+    outputs=gr.Textbox(label="Recommendations"),
+    title="Title Recommendation",
+    description="Do'akan saya langgeng sm Ei"
+)
+
+iface.launch()