{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import numpy as np\nimport pandas as pd\nimport random\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nfrom sklearn.model_selection import train_test_split\nfrom tqdm import tqdm\nprint(\"Setup Done\")","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2021-05-25T03:41:26.303920Z","iopub.execute_input":"2021-05-25T03:41:26.304308Z","iopub.status.idle":"2021-05-25T03:41:27.230297Z","shell.execute_reply.started":"2021-05-25T03:41:26.304274Z","shell.execute_reply":"2021-05-25T03:41:27.229418Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# **Mô tả bài toán**\n\nCho các dữ liệu về các câu hỏi trên Quora dạng text và phân loại của chúng(Có phải toxic hay không?)\n\nĐánh giá xem dữ liệu cho có phải là câu hỏi toxic hay không?","metadata":{}},{"cell_type":"markdown","source":"    Nhập dữ liệu từ file.","metadata":{}},{"cell_type":"code","source":"raw_train_data = pd.read_csv('/kaggle/input/quora-insincere-questions-classification/train.csv')\nraw_test_data = pd.read_csv('/kaggle/input/quora-insincere-questions-classification/test.csv')\nraw_train_data","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:41:27.231538Z","iopub.execute_input":"2021-05-25T03:41:27.231818Z","iopub.status.idle":"2021-05-25T03:41:33.457500Z","shell.execute_reply.started":"2021-05-25T03:41:27.231791Z","shell.execute_reply":"2021-05-25T03:41:33.456614Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Dữ liệu trên tập Train.","metadata":{}},{"cell_type":"code","source":"raw_train_data.head()","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:41:33.459083Z","iopub.execute_input":"2021-05-25T03:41:33.459411Z","iopub.status.idle":"2021-05-25T03:41:33.469631Z","shell.execute_reply.started":"2021-05-25T03:41:33.459385Z","shell.execute_reply":"2021-05-25T03:41:33.468800Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Dữ liệu trên tập Test.","metadata":{}},{"cell_type":"code","source":"raw_test_data.head()","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:41:33.471108Z","iopub.execute_input":"2021-05-25T03:41:33.471496Z","iopub.status.idle":"2021-05-25T03:41:33.481131Z","shell.execute_reply.started":"2021-05-25T03:41:33.471465Z","shell.execute_reply":"2021-05-25T03:41:33.480469Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(len(raw_train_data[raw_train_data.target == 0]))\nprint(len(raw_train_data[raw_train_data.target == 1]))","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:41:33.482483Z","iopub.execute_input":"2021-05-25T03:41:33.482910Z","iopub.status.idle":"2021-05-25T03:41:33.626322Z","shell.execute_reply.started":"2021-05-25T03:41:33.482876Z","shell.execute_reply":"2021-05-25T03:41:33.625624Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> # **Bình luận** :\n\n> **Nhận xét**\n> * Dữ liệu trên tập dữ liệu gồm qid(dùng để test trong cuộc thi),question_text(dữ liệu văn bản của câu hỏi) và target(Có phải câu hỏi toxic hay không)\n> * Chúng ta sẽ không quan sát được dữ liệu như thế nào chỉ từ những câu hỏi dạng text như thế này\n\n> **Vấn đề nảy sinh**\n> * Cần phải đưa dữ liệu câu hỏi dạng text về một dạng có thể quan sát được","metadata":{}},{"cell_type":"markdown","source":"# **Chương 1:Tiền xử lý dữ liệu**\n> Trước tiên,cần xử lý qua dữ liệu text này trước.Những việc cần phải làm những việc sau:\n> * **Quan sát,đánh giá dữ liệu.**\n> * **Xử lý dữ liệu.**","metadata":{}},{"cell_type":"markdown","source":"****","metadata":{}},{"cell_type":"markdown","source":"> # **1.Quan sát dữ liệu**\n> Việc đầu tiên,chúng ta cần xem xét qua dữ liệu đã có nhằm vạch ra một phương hướng xử lý dữ liệu phù hợp.","metadata":{}},{"cell_type":"markdown","source":"> # **1.1-Vấn đề**\n> * Phải đưa dữ liệu về một dạng có thể quan sát được.\n> * Quan sát,đánh giá và đưa ra phương hướng tiền xử lý dữ liệu hiệu quả\n\n","metadata":{}},{"cell_type":"markdown","source":"> # **1.2-Giải pháp**\n> * **Đưa dữ liệu về dạng có thể quan sát được** : Từ một cột dạng text,chúng ta thêm các thuộc tính: độ dài của một câu hỏi,số lượng từ trong một câu hỏi,số lượng các chữ số trong câu,số lượng các kí tự đặc biệt số lượng các từ duy nhất trong câu.\n> * **Quan sát,đánh giá dữ liệu** : Vẽ các biểu đồ dựa vào các thuộc tính mới trên để có một cái nhìn tổng quan nhất về dữ liệu(Sử dụng biểu đồ violin để thấy được mật độ xác suất dữ liệu)","metadata":{}},{"cell_type":"markdown","source":"> > # 1.2.1-Đưa dữ liệu về dạng có thể quan sát được","metadata":{}},{"cell_type":"markdown","source":"    Thêm các thuộc tính mới cho dữ liệu","metadata":{}},{"cell_type":"code","source":"\n    raw_train_data['qlen'] = raw_train_data['question_text'].str.len() \n    raw_train_data['n_words'] = raw_train_data['question_text'].apply(lambda row: len(row.split(\" \")))\n    raw_train_data['numeric_words'] = raw_train_data['question_text'].apply(lambda row: sum(c.isdigit() for c in row))\n    raw_train_data['sp_char_words'] = raw_train_data['question_text'].str.findall(r'[^a-zA-Z0-9 ]').str.len()\n    raw_train_data['unique_words'] = raw_train_data['question_text'].apply(lambda row: len(set(str(row).split())))","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:41:33.627394Z","iopub.execute_input":"2021-05-25T03:41:33.627856Z","iopub.status.idle":"2021-05-25T03:41:54.218475Z","shell.execute_reply.started":"2021-05-25T03:41:33.627825Z","shell.execute_reply":"2021-05-25T03:41:54.217597Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"raw_train_data.head()","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:41:54.219907Z","iopub.execute_input":"2021-05-25T03:41:54.220309Z","iopub.status.idle":"2021-05-25T03:41:54.233109Z","shell.execute_reply.started":"2021-05-25T03:41:54.220269Z","shell.execute_reply":"2021-05-25T03:41:54.231998Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **Nhận xét**: Dữ liệu đã có thể quan sát và trông rõ ràng hơn rất nhiều.","metadata":{}},{"cell_type":"markdown","source":"> > # 1.2.2-Quan sát,đánh giá dữ liệu","metadata":{}},{"cell_type":"markdown","source":"> > **1.2.2.1-Xem số lượng label toxic và label không toxic trên tập dữ liệu**","metadata":{}},{"cell_type":"markdown","source":"> > **Biểu đồ so sánh số lượng hai nhãn trên tập dữ liệu**","metadata":{}},{"cell_type":"code","source":"raw_train_data.groupby(\"target\")['qid'].count().plot.bar()","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:41:54.296996Z","iopub.execute_input":"2021-05-25T03:41:54.297335Z","iopub.status.idle":"2021-05-25T03:41:54.603087Z","shell.execute_reply.started":"2021-05-25T03:41:54.297306Z","shell.execute_reply":"2021-05-25T03:41:54.602392Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Do chỉ có 2 nhãn 0/1 ,ta sử dụng hàm mean() để lấy ra tỉ lệ những câu hỏi không toxic trong toàn bộ tập dữ liệu","metadata":{}},{"cell_type":"code","source":"print('~> Percentage of Sincere Questions (is_duplicate = 0):\\n   {}%'.format(100 - round(raw_train_data['target'].mean()*100, 2)))\nprint('\\n~> Percentage of Insincere Questions (is_duplicate = 1):\\n   {}%'.format(round(raw_train_data['target'].mean()*100, 2)))","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:41:54.604741Z","iopub.execute_input":"2021-05-25T03:41:54.605210Z","iopub.status.idle":"2021-05-25T03:41:54.614973Z","shell.execute_reply.started":"2021-05-25T03:41:54.605178Z","shell.execute_reply":"2021-05-25T03:41:54.613994Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **Nhận xét** : Tỉ lệ của các câu hỏi trung thực là 93,81%,các câu hỏi không trung thực là 6.19% trên toàn bộ tập train cho thấy sự chênh lệch quá lớn giữa 2 nhãn.Việc chênh lệch này cực kỳ không tốt,có thể làm cho model sau này hoạt động kém hiệu quả","metadata":{}},{"cell_type":"markdown","source":"> > **1.2.2.2-Sự phân bố về số lượng từ trong các câu ở 2 nhãn trên tập dữ liệu**","metadata":{}},{"cell_type":"markdown","source":"> > **Biểu đồ phân bố số lượng từ**","metadata":{}},{"cell_type":"code","source":"plt.figure(figsize=(8, 16))\nsns.violinplot(x = 'target', y = 'n_words', data = raw_train_data[0:])\nplt.yticks(ticks=[0,5,10,15,20,25,30,35,40,45,50,55,60,80,100,120])\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:41:54.616152Z","iopub.execute_input":"2021-05-25T03:41:54.616434Z","iopub.status.idle":"2021-05-25T03:41:57.689077Z","shell.execute_reply.started":"2021-05-25T03:41:54.616408Z","shell.execute_reply":"2021-05-25T03:41:57.688338Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **Nhận xét**:\n> > * **Câu hỏi không toxic**:\n     * số lượng từ phần lớn giao động trong khoảng 2-30 từ,số lượng từ nhỏ nhất trong một câu là 1 từ,lớn nhất là khoảng 120 từ\n> > * **Câu hỏi toxic**:\n     * số lượng từ  phần lớn giao động trong khoảng 2-40 từ,số lượng từ nhỏ nhất là 1 từ,lớn nhất là khoảng 70 từ","metadata":{}},{"cell_type":"markdown","source":"> > **1.2.2.3-Sự phân bố về số lượng chữ số trong các câu ở 2 nhãn trên tập dữ liệu**","metadata":{}},{"cell_type":"markdown","source":"> > **Biểu đồ phân bố các chữ số**","metadata":{}},{"cell_type":"code","source":"plt.figure(figsize=(8, 20))\nsns.violinplot(x = 'target', y = 'numeric_words', data = raw_train_data[0:])\nplt.yticks(ticks=[0,2,4,6,8,10,50,100,150,200])\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:41:57.690123Z","iopub.execute_input":"2021-05-25T03:41:57.690506Z","iopub.status.idle":"2021-05-25T03:42:00.329019Z","shell.execute_reply.started":"2021-05-25T03:41:57.690478Z","shell.execute_reply":"2021-05-25T03:42:00.327198Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **Nhận xét**:\n> > * **Câu hỏi không toxic**:\n     * số lượng chữ số phần lớn giao động trong khoảng 0-4 từ,số lượng chữ số nhỏ nhất trong một câu là 0 chữ số,lớn nhất là khoảng 230 chữ số\n> > * **Câu hỏi toxic**:\n     * số lượng chữ số phần lớn giao động trong khoảng 0-1 từ,số lượng chữ số nhỏ nhất là 0 chữ số,lớn nhất là khoảng 90 chữ số\n\n> > => câu hỏi toxic có vẻ ít chữ số hơn câu hỏi không toxic","metadata":{}},{"cell_type":"markdown","source":"> > **1.2.2.3-Sự phân bố về số lượng kí tự đặc biệt trong các câu ở 2 nhãn trên tập dữ liệu**","metadata":{}},{"cell_type":"markdown","source":"> > **Biểu đồ phân bố các kí tự đặc biệt**","metadata":{}},{"cell_type":"code","source":"plt.figure(figsize=(8, 20))\nsns.violinplot(x = 'target', y = 'sp_char_words', data = raw_train_data[0:])\nplt.yticks(ticks=[0,5,10,20,30,40,50,100,200,300,400])\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:00.330381Z","iopub.execute_input":"2021-05-25T03:42:00.330795Z","iopub.status.idle":"2021-05-25T03:42:03.004830Z","shell.execute_reply.started":"2021-05-25T03:42:00.330752Z","shell.execute_reply":"2021-05-25T03:42:03.003782Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **Nhận xét**:\n> > * **Câu hỏi không toxic**:\n     * số lượng kí tự đặc biệt phần lớn giao động trong khoảng 0-4 từ,số lượng kí tự đặc biệt nhỏ nhất là 0,lớn nhất là khoảng 170\n> > * **Câu hỏi toxic**:\n     * số lượng kí tự đặc biệt phần lớn giao động trong khoảng 0-7 từ,số lượng kí tự đặc biệt nhỏ nhất là 0,lớn nhất là khoảng 420\n     \n> > => có thể thấy được số lượng kí tự đặc biệt trong câu hỏi toxic,nhiều hơn nhiều so với câu hỏi không toxic","metadata":{}},{"cell_type":"markdown","source":"> > **1.2.2.4-Sự phân bố về số lượng từ duy nhất trong các câu ở 2 nhãn trên tập dữ liệu**","metadata":{}},{"cell_type":"markdown","source":"> > **Biểu đồ về sự phân bố các từ duy nhất**","metadata":{}},{"cell_type":"code","source":"plt.figure(figsize=(8, 16))\nsns.violinplot(x = 'target', y = 'unique_words', data = raw_train_data[0:])\nplt.yticks(ticks=[0,2,4,6,8,10,20,30,40,50,60,80,100])\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:03.006224Z","iopub.execute_input":"2021-05-25T03:42:03.006601Z","iopub.status.idle":"2021-05-25T03:42:06.094756Z","shell.execute_reply.started":"2021-05-25T03:42:03.006571Z","shell.execute_reply":"2021-05-25T03:42:06.093638Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **Nhận xét**:\n> > * **Câu hỏi không toxic**:\n     * số lượng từ duy nhất phần lớn giao động trong khoảng 3-20 từ,số lượng từ duy nhất nhỏ nhất là 2,lớn nhất là khoảng 90\n> > * **Câu hỏi toxic**:\n     * số lượng từ duy nhất phần lớn giao động trong khoảng 2-30 từ,số lượng từ duy nhất nhỏ nhất là 2,lớn nhất là khoảng 50\n     \n> > => số lượng từ duy nhất trong câu hỏi toxic có phân bố dàn trải hơn câu hỏi không toxic trên tập dữ liệu","metadata":{}},{"cell_type":"markdown","source":"> # **1.3-Kết quả**\n> Dữ liệu đã dễ quan sát hơn rất nhiều và có thể thu thập được nhiều thông tin hữu ích hơn,hoàn toàn có thể đưa ra hướng xử lý dữ liệu thô tốt","metadata":{}},{"cell_type":"markdown","source":"> # **1.4-Bình luận** :\n>  * **Nhận xét**:\n>   * **Tỉ lệ các nhãn trên tập dữ liệu**: Tỉ lệ giữa câu hỏi toxic và không toxic chênh lệch rất lớn(không toxic:93.81%,toxic:6,19%) dễ dàng có thể gây ảnh hưởng tới độ chính xác và hiệu quả của model sau này.\n>   * **Dữ liệu có nhiều nhiễu**:Từ biểu đồ dữ liệu ở các cột số,kí tự đặc biệt có thể thấy rằng dữ liệu có rất nhiều nhiễu.\n>  * **Vấn đề nảy sinh**:\n>   * Cần cân bằng lại số lượng label trên tập dữ liệu.\n>   * Cần loại bỏ bớt các nhiễu của dữ liệu\n>  * **Ý kiến**:Còn thiếu dữ liệu quan sát cho các từ viết tắt","metadata":{}},{"cell_type":"markdown","source":"    ","metadata":{}},{"cell_type":"markdown","source":"> # **2.Xử lý dữ liệu**\n> Khi xử lý dữ liệu,cần lọc ra những dữ liệu cần thiết và không cần thiết cho việc dự đoán đánh giá của model,chia lại tập dữ liệu để cân bằng về mặt số lượng giữa 2 label và chia tập train,validation.","metadata":{}},{"cell_type":"markdown","source":"> # **2.1-Vấn đề**\n> * **Cân bằng lại số lượng label cho tập dữ liệu**\n> * **Loại bỏ nhiễu**","metadata":{}},{"cell_type":"markdown","source":"> # **2.2-Cách giải quyết**\n> * **Cân bằng lại số lượng label cho tập dữ liệu**:Sử dụng resample để chia lại tập dữ liệu sao cho cân bằng về số lượng 2 nhãn hơn\n> * **Loại bỏ nhiễu**:Sử dụng một hàm để lọc từ viết tắt và thư việc re để lọc kí tự đặc biệt + chữ số","metadata":{}},{"cell_type":"markdown","source":"> > # 2.2.1-Cân bằng lại số lượng label cho tập dữ liệu","metadata":{}},{"cell_type":"markdown","source":"> > **2.2.1.1-Chia lại tập dữ liệu theo tỉ lệ 3:1 bằng resample**","metadata":{}},{"cell_type":"code","source":"#Chia lại tập dữ liệu theo tỉ lệ 3:1\nfrom sklearn.utils import resample\nsincere = raw_train_data[raw_train_data.target == 0]\ninsincere = raw_train_data[raw_train_data.target == 1]\nraw_train_data = pd.concat([resample(sincere,replace = True,n_samples = len(insincere)*3), insincere])\nprint(len(raw_train_data[raw_train_data.target == 0]))\nprint(len(raw_train_data[raw_train_data.target == 1]))","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:06.096036Z","iopub.execute_input":"2021-05-25T03:42:06.096343Z","iopub.status.idle":"2021-05-25T03:42:06.476505Z","shell.execute_reply.started":"2021-05-25T03:42:06.096313Z","shell.execute_reply":"2021-05-25T03:42:06.475804Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **2.2.2.2-Chia tập train và tập validation**\n> > Phải tách tập train và tập validation ra sớm,tránh tập validation có thể bị ô nhiễm bởi tập train trong quá trình tiền xử lý dữ liệu","metadata":{}},{"cell_type":"markdown","source":"    Chia dữ liệu ra tập traning và tập validation","metadata":{}},{"cell_type":"code","source":"### chia ra thành 2 tập,tập train và tập validation\nfeature_name = ['question_text','target']\ntrain_data = raw_train_data[feature_name]\ntrain,val=train_test_split(train_data,test_size=0.3,stratify=train_data.target,random_state=123)\nprint(\"Shape of the Training set :\",train.shape)\nprint(\"Shape of the Validation set :\",val.shape)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:06.477745Z","iopub.execute_input":"2021-05-25T03:42:06.478219Z","iopub.status.idle":"2021-05-25T03:42:06.738801Z","shell.execute_reply.started":"2021-05-25T03:42:06.478187Z","shell.execute_reply":"2021-05-25T03:42:06.737819Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Chúng ta phải reset_index cho dữ liệu mới để đồng bộ.","metadata":{}},{"cell_type":"code","source":"train = train.reset_index()\nval = val.reset_index()\nprint(train.head())","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:06.740084Z","iopub.execute_input":"2021-05-25T03:42:06.740362Z","iopub.status.idle":"2021-05-25T03:42:06.769686Z","shell.execute_reply.started":"2021-05-25T03:42:06.740335Z","shell.execute_reply":"2021-05-25T03:42:06.768500Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > # 2.2.2-Loại bỏ các từ viết tắt","metadata":{}},{"cell_type":"markdown","source":"> > **2.2.2.1 Loại bỏ các từ viết tắt**","metadata":{}},{"cell_type":"code","source":"###Loại bỏ các từ viết tắt \n\n#Tạo một directory có key là các từ viết tắt và value là các từ không viết tắt của key\n#Mục đích để thay thế các từ viết tắt bằng các từ chuẩn,để chuẩn bị tạo danh sách từ vựng\ncontractions={\"I'm\": 'I am',\n \"I'm'a\": 'I am about to',\n \"I'm'o\": 'I am going to',\n \"I've\": 'I have',\n \"I'll\": 'I will',\n \"I'll've\": 'I will have',\n \"I'd\": 'I would',\n \"I'd've\": 'I would have',\n 'Whatcha': 'What are you',\n \"amn't\": 'am not',\n \"ain't\": 'are not',\n \"aren't\": 'are not',\n \"'cause\": 'because',\n \"can't\": 'can not',\n \"can't've\": 'can not have',\n \"could've\": 'could have',\n \"couldn't\": 'could not',\n \"couldn't've\": 'could not have',\n \"daren't\": 'dare not',\n \"daresn't\": 'dare not',\n \"dasn't\": 'dare not',\n \"didn't\": 'did not',\n 'didn’t': 'did not',\n \"don't\": 'do not',\n 'don’t': 'do not',\n \"doesn't\": 'does not',\n \"e'er\": 'ever',\n \"everyone's\": 'everyone is',\n 'finna': 'fixing to',\n 'gimme': 'give me',\n \"gon't\": 'go not',\n 'gonna': 'going to',\n 'gotta': 'got to',\n \"hadn't\": 'had not',\n \"hadn't've\": 'had not have',\n \"hasn't\": 'has not',\n \"haven't\": 'have not',\n \"he've\": 'he have',\n \"he's\": 'he is',\n \"he'll\": 'he will',\n \"he'll've\": 'he will have',\n \"he'd\": 'he would',\n \"he'd've\": 'he would have',\n \"here's\": 'here is',\n \"how're\": 'how are',\n \"how'd\": 'how did',\n \"how'd'y\": 'how do you',\n \"how's\": 'how is',\n \"how'll\": 'how will',\n \"isn't\": 'is not',\n \"it's\": 'it is',\n \"'tis\": 'it is',\n \"'twas\": 'it was',\n \"it'll\": 'it will',\n \"it'll've\": 'it will have',\n \"it'd\": 'it would',\n \"it'd've\": 'it would have',\n 'kinda': 'kind of',\n \"let's\": 'let us',\n 'luv': 'love',\n \"ma'am\": 'madam',\n \"may've\": 'may have',\n \"mayn't\": 'may not',\n \"might've\": 'might have',\n \"mightn't\": 'might not',\n \"mightn't've\": 'might not have',\n \"must've\": 'must have',\n \"mustn't\": 'must not',\n \"mustn't've\": 'must not have',\n \"needn't\": 'need not',\n \"needn't've\": 'need not have',\n \"ne'er\": 'never',\n \"o'\": 'of',\n \"o'clock\": 'of the clock',\n \"ol'\": 'old',\n \"oughtn't\": 'ought not',\n \"oughtn't've\": 'ought not have',\n \"o'er\": 'over',\n \"shan't\": 'shall not',\n \"sha'n't\": 'shall not',\n \"shalln't\": 'shall not',\n \"shan't've\": 'shall not have',\n \"she's\": 'she is',\n \"she'll\": 'she will',\n \"she'd\": 'she would',\n \"she'd've\": 'she would have',\n \"should've\": 'should have',\n \"shouldn't\": 'should not',\n \"shouldn't've\": 'should not have',\n \"so've\": 'so have',\n \"so's\": 'so is',\n \"somebody's\": 'somebody is',\n \"someone's\": 'someone is',\n \"something's\": 'something is',\n 'sux': 'sucks',\n \"that're\": 'that are',\n \"that's\": 'that is',\n \"that'll\": 'that will',\n \"that'd\": 'that would',\n \"that'd've\": 'that would have',\n 'em': 'them',\n \"there're\": 'there are',\n \"there's\": 'there is',\n \"there'll\": 'there will',\n \"there'd\": 'there would',\n \"there'd've\": 'there would have',\n \"these're\": 'these are',\n \"they're\": 'they are',\n \"they've\": 'they have',\n \"they'll\": 'they will',\n \"they'll've\": 'they will have',\n \"they'd\": 'they would',\n \"they'd've\": 'they would have',\n \"this's\": 'this is',\n \"those're\": 'those are',\n \"to've\": 'to have',\n 'wanna': 'want to',\n \"wasn't\": 'was not',\n \"we're\": 'we are',\n \"we've\": 'we have',\n \"we'll\": 'we will',\n \"we'll've\": 'we will have',\n \"we'd\": 'we would',\n \"we'd've\": 'we would have',\n \"weren't\": 'were not',\n \"what're\": 'what are',\n \"what'd\": 'what did',\n \"what've\": 'what have',\n \"what's\": 'what is',\n \"what'll\": 'what will',\n \"what'll've\": 'what will have',\n \"when've\": 'when have',\n \"when's\": 'when is',\n \"where're\": 'where are',\n \"where'd\": 'where did',\n \"where've\": 'where have',\n \"where's\": 'where is',\n \"which's\": 'which is',\n \"who're\": 'who are',\n \"who've\": 'who have',\n \"who's\": 'who is',\n \"who'll\": 'who will',\n \"who'll've\": 'who will have',\n \"who'd\": 'who would',\n \"who'd've\": 'who would have',\n \"why're\": 'why are',\n \"why'd\": 'why did',\n \"why've\": 'why have',\n \"why's\": 'why is',\n \"will've\": 'will have',\n \"won't\": 'will not',\n \"won't've\": 'will not have',\n \"would've\": 'would have',\n \"wouldn't\": 'would not',\n \"wouldn't've\": 'would not have',\n \"y'all\": 'you all',\n \"y'all're\": 'you all are',\n \"y'all've\": 'you all have',\n \"y'all'd\": 'you all would',\n \"y'all'd've\": 'you all would have',\n \"you're\": 'you are',\n \"you've\": 'you have',\n \"you'll've\": 'you shall have',\n \"you'll\": 'you will',\n \"you'd\": 'you would',\n \"you'd've\": 'you would have',\n 'jan.': 'january',\n 'feb.': 'february',\n 'mar.': 'march',\n 'apr.': 'april',\n 'jun.': 'june',\n 'jul.': 'july',\n 'aug.': 'august',\n 'sep.': 'september',\n 'oct.': 'october',\n 'nov.': 'november',\n 'dec.': 'december',\n 'I’m': 'I am',\n 'I’m’a': 'I am about to',\n 'I’m’o': 'I am going to',\n 'I’ve': 'I have',\n 'I’ll': 'I will',\n 'I’ll’ve': 'I will have',\n 'I’d': 'I would',\n 'I’d’ve': 'I would have',\n 'amn’t': 'am not',\n 'ain’t': 'are not',\n 'aren’t': 'are not',\n '’cause': 'because',\n 'can’t': 'can not',\n 'can’t’ve': 'can not have',\n 'could’ve': 'could have',\n 'couldn’t': 'could not',\n 'couldn’t’ve': 'could not have',\n 'daren’t': 'dare not',\n 'daresn’t': 'dare not',\n 'dasn’t': 'dare not',\n 'doesn’t': 'does not',\n 'e’er': 'ever',\n 'everyone’s': 'everyone is',\n 'gon’t': 'go not',\n 'hadn’t': 'had not',\n 'hadn’t’ve': 'had not have',\n 'hasn’t': 'has not',\n 'haven’t': 'have not',\n 'he’ve': 'he have',\n 'he’s': 'he is',\n 'he’ll': 'he will',\n 'he’ll’ve': 'he will have',\n 'he’d': 'he would',\n 'he’d’ve': 'he would have',\n 'here’s': 'here is',\n 'how’re': 'how are',\n 'how’d': 'how did',\n 'how’d’y': 'how do you',\n 'how’s': 'how is',\n 'how’ll': 'how will',\n 'isn’t': 'is not',\n 'it’s': 'it is',\n '’tis': 'it is',\n '’twas': 'it was',\n 'it’ll': 'it will',\n 'it’ll’ve': 'it will have',\n 'it’d': 'it would',\n 'it’d’ve': 'it would have',\n 'let’s': 'let us',\n 'ma’am': 'madam',\n 'may’ve': 'may have',\n 'mayn’t': 'may not',\n 'might’ve': 'might have',\n 'mightn’t': 'might not',\n 'mightn’t’ve': 'might not have',\n 'must’ve': 'must have',\n 'mustn’t': 'must not',\n 'mustn’t’ve': 'must not have',\n 'needn’t': 'need not',\n 'needn’t’ve': 'need not have',\n 'ne’er': 'never',\n 'o’': 'of',\n 'o’clock': 'of the clock',\n 'ol’': 'old',\n 'oughtn’t': 'ought not',\n 'oughtn’t’ve': 'ought not have',\n 'o’er': 'over',\n 'shan’t': 'shall not',\n 'sha’n’t': 'shall not',\n 'shalln’t': 'shall not',\n 'shan’t’ve': 'shall not have',\n 'she’s': 'she is',\n 'she’ll': 'she will',\n 'she’d': 'she would',\n 'she’d’ve': 'she would have',\n 'should’ve': 'should have',\n 'shouldn’t': 'should not',\n 'shouldn’t’ve': 'should not have',\n 'so’ve': 'so have',\n 'so’s': 'so is',\n 'somebody’s': 'somebody is',\n 'someone’s': 'someone is',\n 'something’s': 'something is',\n 'that’re': 'that are',\n 'that’s': 'that is',\n 'that’ll': 'that will',\n 'that’d': 'that would',\n 'that’d’ve': 'that would have',\n 'there’re': 'there are',\n 'there’s': 'there is',\n 'there’ll': 'there will',\n 'there’d': 'there would',\n 'there’d’ve': 'there would have',\n 'these’re': 'these are',\n 'they’re': 'they are',\n 'they’ve': 'they have',\n 'they’ll': 'they will',\n 'they’ll’ve': 'they will have',\n 'they’d': 'they would',\n 'they’d’ve': 'they would have',\n 'this’s': 'this is',\n 'those’re': 'those are',\n 'to’ve': 'to have',\n 'wasn’t': 'was not',\n 'we’re': 'we are',\n 'we’ve': 'we have',\n 'we’ll': 'we will',\n 'we’ll’ve': 'we will have',\n 'we’d': 'we would',\n 'we’d’ve': 'we would have',\n 'weren’t': 'were not',\n 'what’re': 'what are',\n 'what’d': 'what did',\n 'what’ve': 'what have',\n 'what’s': 'what is',\n 'what’ll': 'what will',\n 'what’ll’ve': 'what will have',\n 'when’ve': 'when have',\n 'when’s': 'when is',\n 'where’re': 'where are',\n 'where’d': 'where did',\n 'where’ve': 'where have',\n 'where’s': 'where is',\n 'which’s': 'which is',\n 'who’re': 'who are',\n 'who’ve': 'who have',\n 'who’s': 'who is',\n 'who’ll': 'who will',\n 'who’ll’ve': 'who will have',\n 'who’d': 'who would',\n 'who’d’ve': 'who would have',\n 'why’re': 'why are',\n 'why’d': 'why did',\n 'why’ve': 'why have',\n 'why’s': 'why is',\n 'will’ve': 'will have',\n 'won’t': 'will not',\n 'won’t’ve': 'will not have',\n 'would’ve': 'would have',\n 'wouldn’t': 'would not',\n 'wouldn’t’ve': 'would not have',\n 'y’all': 'you all',\n 'y’all’re': 'you all are',\n 'y’all’ve': 'you all have',\n 'y’all’d': 'you all would',\n 'y’all’d’ve': 'you all would have',\n 'you’re': 'you are',\n 'you’ve': 'you have',\n 'you’ll’ve': 'you shall have',\n 'you’ll': 'you will',\n 'you’d': 'you would',\n 'you’d’ve': 'you would have'}\n\n#Hàm chuyển đổi các từ viết tắt thành các cụm từ chuẩn,không viết tắt\ndef contraction_fix(word):\n    try:\n        a=contractions[word]#nếu word là từ viết tắt có trong bộ từ viết tắt => a sẽ là cụm từ không viết tắt của word\n    except KeyError:\n        a=word # nếu không có key nào trong directory phù hợp với word đã cho=> a sẽ vẫn là word\n    return a #trả về từ vựng(Cụm từ vựng không viết tắt)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:06.771122Z","iopub.execute_input":"2021-05-25T03:42:06.771407Z","iopub.status.idle":"2021-05-25T03:42:06.807933Z","shell.execute_reply.started":"2021-05-25T03:42:06.771380Z","shell.execute_reply":"2021-05-25T03:42:06.806759Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **2.2.2.2 Loại bỏ các số,kí tự đặc biệt**","metadata":{}},{"cell_type":"code","source":"###Loại bỏ các chữ số,các kí tự đặc biệt\nimport re\ndef Preprocess(doc): #Hàm loại bỏ các chữ số,các kí tự đặc biệt\n    corpus=[]\n    for text in tqdm(doc):\n        text=\" \".join([contraction_fix(w) for w in text.split()])   #tách các từ,thay thế các từ viết tắt bằng các từ đúng,sau đó nối chúng lại\n        \n        #re là một module để xác định biểu thức chính quy\n        text=re.sub(r'[^a-z0-9A-Z]',\" \",text)#Loại bỏ các dấu như !,?.... thay bằng các ' '\n        text=re.sub(r'[0-9]{1}',\"#\",text)#Loại bỏ các số,thay bằng các '#'\n        text=re.sub(r'[0-9]{2}','##',text)\n        text=re.sub(r'[0-9]{3}','###',text)\n        text=re.sub(r'[0-9]{4}','####',text)\n        text=re.sub(r'[0-9]{5,}','#####',text)\n        corpus.append(text) #thêm dòng vừa rồi vào mảng corpus\n    return corpus #trả về mảng các câu đã được lược bỏ số và các kí tự đặc biệt","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:06.809508Z","iopub.execute_input":"2021-05-25T03:42:06.809975Z","iopub.status.idle":"2021-05-25T03:42:06.825572Z","shell.execute_reply.started":"2021-05-25T03:42:06.809931Z","shell.execute_reply":"2021-05-25T03:42:06.824572Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Xử lý,chuyển đổi dữ liệu thô sang dữ liệu đã được sơ chế,loại bỏ nhiễu","metadata":{}},{"cell_type":"code","source":"train_processed_doc = Preprocess(train.question_text)\nval_processed_doc = Preprocess(val.question_text)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:06.826938Z","iopub.execute_input":"2021-05-25T03:42:06.827490Z","iopub.status.idle":"2021-05-25T03:42:19.684885Z","shell.execute_reply.started":"2021-05-25T03:42:06.827454Z","shell.execute_reply":"2021-05-25T03:42:19.683866Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Kiểm tra xem hàm preprocess đã hoạt động đúng chưa","metadata":{}},{"cell_type":"code","source":"for i in range(0,10,1):\n    randIndex = random.randrange(0,train.question_text.size,1)\n    print(\"Raw:\" + train.question_text[randIndex])\n    print(\"Process:\" + train_processed_doc[randIndex])","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:19.686321Z","iopub.execute_input":"2021-05-25T03:42:19.686724Z","iopub.status.idle":"2021-05-25T03:42:19.693923Z","shell.execute_reply.started":"2021-05-25T03:42:19.686671Z","shell.execute_reply":"2021-05-25T03:42:19.693231Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> # **2.3-Kết quả** \n> Số lượng nhãn trong tập train đã cân bằng hơn;các số,từ viết tắt,chữ đã được loại bỏ.Dữ liệu bây giờ sẽ tốt hơn cho mô hình của chúng ta","metadata":{}},{"cell_type":"markdown","source":"> # **2.4-Bình luận** :\n>  * **Nhận xét**:\n>   * **Hàm lọc từ viết tắt**: Dữ liệu các từ viết tắt của hàm là thu thập bằng tay,vì thế có khả năng sẽ có những từ viết tắt không được thay đổi,không biết có một thư viện nào chuyên lọc những từ viết tắt như thế này không?\n>   * **Cân bằng label trên tập dữ liệu**:Resample bằng cách undersampling nên có một lượng lớn dữ liệu các câu hỏi không toxic bị mất đi.\n>  * **Ý kiến**:Tôi sẽ thử thêm oversampling với dữ liệu sau","metadata":{}},{"cell_type":"markdown","source":"# **Chương 2:Trích chọn đặc trưng từ dữ liệu,tạo vector dữ liệu**\n> \n> Khi đã xử lý qua được dữ liệu,chúng ta cần tiếp tục chuyển đổi dữ liệu vừa được sơ chế đó sang dữ liệu có thể training cho model\n> * **Tạo Vector từ dữ liệu bằng cách thủ công**\n> * **Tạo Vector từ dữ liệu bằng tf-idf**","metadata":{}},{"cell_type":"markdown","source":"> # **1.Tạo Vector từ dữ liệu bằng cách thủ công**\n","metadata":{}},{"cell_type":"markdown","source":"> # **1.1-Vấn đề** \n> Làm sao để lấy ra được từ vựng và số lượng của mỗi từ vựng từ văn bản đã qua xử lý,sau đó kiến tạo vector feature từ những từ vựng đã thu được đó.","metadata":{}},{"cell_type":"markdown","source":"> # **1.2-Cách giải quyết** \n>  * **Lấy ra từ vựng** : Khởi tạo một dictionary rỗng,lặp qua các câu,dùng thư viện của python để tách từ trong các câu.Mỗi khi quét đến một từ nào đó,kiểm tra xem từ đó có trong dictionary từ vựng hay không.Nếu có,cộng thêm 1 vào value của từ đó trong directory(từ đó là key của phần tử trong dictionary).Nếu không,khởi tạo một phần tử mới cho directory với key = từ đó và value = 1.\n>  * **Kiến tạo vector feature**:\n>    * Phương án 1:Tạo vector đặc từ toàn bộ từ vựng đã có.\n>    * Phương án 2:Tạo vector thưa từ toàn bộ từ vựng đã có.","metadata":{}},{"cell_type":"markdown","source":"> > # 1.2.1-Lấy ra từ vựng cho bag of word","metadata":{}},{"cell_type":"markdown","source":"    Hàm lấy từ vựng từ các đoạn văn bản","metadata":{}},{"cell_type":"code","source":"###Sau khi đã có hàm tiền xử lý cần thiết,tạo hàm lấy ra vốn từ vựng\n\ndef get_vocab(corpus):\n    vocab={}#Đây là vốn từ vựng của chúng ta(Directory)\n    for text in tqdm(corpus): #Lặp qua các câu trong danh sách câu đã được xử lý\n        for word in text.split(): #Lặp qua các từ trong các câu\n            try:\n                vocab[word]+=1 #Nếu từ đó đã có rồi,+1 thêm vào số lượng của từ đó\n            except KeyError:\n                vocab[word]=1 #Nếu chưa có từ đó,tạo ra 1 key = word và value = 1 mới trong vocab(Directory)\n    vocab=dict(sorted(vocab.items(),reverse=True ,key=lambda item: item[1]))#Sắp kếp lại các Key\n    return vocab  # Hàm trả về một dictionary có các key là toàn bộ từ vựng của văn bản \n                  #và value là số lượng từ đó xuất hiện trong toàn bộ văn bản","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:19.694806Z","iopub.execute_input":"2021-05-25T03:42:19.695058Z","iopub.status.idle":"2021-05-25T03:42:19.705416Z","shell.execute_reply.started":"2021-05-25T03:42:19.695032Z","shell.execute_reply":"2021-05-25T03:42:19.704417Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Tiến hành lấy ra các từ vựng,và số lượng của các từ trong văn bản","metadata":{}},{"cell_type":"code","source":"vocabulary = get_vocab(train_processed_doc)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:19.707152Z","iopub.execute_input":"2021-05-25T03:42:19.707617Z","iopub.status.idle":"2021-05-25T03:42:20.860815Z","shell.execute_reply.started":"2021-05-25T03:42:19.707571Z","shell.execute_reply":"2021-05-25T03:42:20.859849Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > # 1.2.2-Quan sát,đánh giá dữ liệu\n   ","metadata":{}},{"cell_type":"markdown","source":"> > **1.2.2.1-Miêu tả dữ liệu**","metadata":{}},{"cell_type":"markdown","source":"    Do dictionary đã tự động sắp xếp dữ liệu theo values từ lớn đến bé,nên chúng ta không cần sắp xếp lại nữa.\n    Tạo dataframe để quan sát","metadata":{}},{"cell_type":"code","source":"diagram_x = {'num_of_word' : list(vocabulary.values())}\ndf_index = list(vocabulary.keys())\ndf = pd.DataFrame(diagram_x,index=df_index)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:20.866431Z","iopub.execute_input":"2021-05-25T03:42:20.866768Z","iopub.status.idle":"2021-05-25T03:42:20.934111Z","shell.execute_reply.started":"2021-05-25T03:42:20.866735Z","shell.execute_reply":"2021-05-25T03:42:20.933393Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Những từ xuất hiện nhiều nhất.","metadata":{}},{"cell_type":"code","source":"df[0 : 10]","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:20.936340Z","iopub.execute_input":"2021-05-25T03:42:20.936628Z","iopub.status.idle":"2021-05-25T03:42:20.950755Z","shell.execute_reply.started":"2021-05-25T03:42:20.936601Z","shell.execute_reply":"2021-05-25T03:42:20.949466Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Những từ xuất hiện ít nhất","metadata":{}},{"cell_type":"code","source":"df[-10:-1]","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:20.952438Z","iopub.execute_input":"2021-05-25T03:42:20.953083Z","iopub.status.idle":"2021-05-25T03:42:20.966426Z","shell.execute_reply.started":"2021-05-25T03:42:20.953038Z","shell.execute_reply":"2021-05-25T03:42:20.965399Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **1.2.2.2-Nhận xét**:\n> > * Những từ có số lượng nhiều nhất là những từ thông thường hay gặp trong nhiều loại câu,ví dụ như từ 'the',có số lượng từ nhiều nhất vì nó là từ có thể gặp trong rất nhiều thể loại câu => gần như nó không có tác dụng đặc trưng cho 1 loại câu cụ thể.\n> > * Những từ có số lượng ít nhất có thể là những từ sai chính tả,hoặc tên riêng => những từ như vậy cũng không có tác dụng đặc trưng cho một loại câu cụ thể.","metadata":{}},{"cell_type":"markdown","source":"> > # 1.2.3-Tạo vector feature","metadata":{}},{"cell_type":"markdown","source":"> > **1.2.2.1-Dùng ma trận đặc**","metadata":{}},{"cell_type":"markdown","source":"    Hàm này sẽ tạo ra một mảng 2 chiều,mỗi phần tử của mảng là một câu,mỗi phần tử của phần tử của mảng là số lần từ tương ứng với vị trí(trong từ vựng) xuất hiện trong câu","metadata":{}},{"cell_type":"code","source":"##thử xử lý dữ liệu theo kiểu nguyên thủy nhất :))))\ndef naiveProcessData(raw_data):\n    processedData = []\n    for data in tqdm(raw_data):\n        #tạo mới directory với các key là vocabulary và value = 0;\n        processedSentence = vocabulary.copy()\n        for key in dictOfVocabulary.keys():\n            processedSentence[key] = 0\n        \n        #cộng 1 với mỗi word có trong câu.\n        for word in data.split():\n            try:\n                processedSentence[word] += 1;    \n            except:\n                print(word + \"is not in bag of word\")\n        processedData.append(np.array(processedSentence.values()))\n    return processedData","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:20.967666Z","iopub.execute_input":"2021-05-25T03:42:20.967972Z","iopub.status.idle":"2021-05-25T03:42:20.974139Z","shell.execute_reply.started":"2021-05-25T03:42:20.967945Z","shell.execute_reply":"2021-05-25T03:42:20.973036Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"### Test\n#processedSentence = vocabulary.copy()\n#for key in dictOfVocabulary.keys():\n#   processedSentence[key] = 0\n#print(np.array(processedSentence.values()))","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:20.975303Z","iopub.execute_input":"2021-05-25T03:42:20.975584Z","iopub.status.idle":"2021-05-25T03:42:20.989896Z","shell.execute_reply.started":"2021-05-25T03:42:20.975558Z","shell.execute_reply":"2021-05-25T03:42:20.988967Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Thử xử lý dữ liệu bằng cách vector hóa các câu sử dụng ma trận thường.","metadata":{}},{"cell_type":"code","source":"#processedData = naiveProcessData(train_processed_doc);","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:20.991642Z","iopub.execute_input":"2021-05-25T03:42:20.992140Z","iopub.status.idle":"2021-05-25T03:42:21.001395Z","shell.execute_reply.started":"2021-05-25T03:42:20.992087Z","shell.execute_reply":"2021-05-25T03:42:21.000694Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **Nhận xét**: Cách sử dụng ma trận đặc không hề ổn chút nào,dữ liệu quá lớn dẫn đến còn không đủ bộ nhớ để lưu trữ ,không thể xử lý dữ liệu theo cách này.","metadata":{}},{"cell_type":"markdown","source":"> > **1.2.2.2-Dùng ma trận rỗng**","metadata":{}},{"cell_type":"markdown","source":"    Hàm này nhằm đánh số thứ tự cho các từ","metadata":{}},{"cell_type":"code","source":"###Tạo hàm khởi tạo các giá trị cần thiết cho ma trận thưa\n#Đầu tiên,phải đánh dấu vị trí của các từ trước\nindex_Vocabulary = vocabulary.copy()\ndef initIndexVocabulary(indexVocabulary):\n    i = 0;\n    for key in indexVocabulary.keys():\n        indexVocabulary[key] = i\n        i += 1","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:21.002291Z","iopub.execute_input":"2021-05-25T03:42:21.002568Z","iopub.status.idle":"2021-05-25T03:42:21.016259Z","shell.execute_reply.started":"2021-05-25T03:42:21.002541Z","shell.execute_reply":"2021-05-25T03:42:21.015317Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Gọi hàm khởi tạo để đánh số thứ tự cho từng từ trong vocabulary(Tôi sẽ dùng tất cả từ vựng tìm được ở tập traning + 1 cột cho những từ không xuất hiện trong tập traning làm feature cho vector)","metadata":{}},{"cell_type":"code","source":"initIndexVocabulary(index_Vocabulary)\n#print(index_Vocabulary.items())","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:21.017334Z","iopub.execute_input":"2021-05-25T03:42:21.017595Z","iopub.status.idle":"2021-05-25T03:42:21.050273Z","shell.execute_reply.started":"2021-05-25T03:42:21.017569Z","shell.execute_reply":"2021-05-25T03:42:21.049361Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Hàm này nhằm mục đích lấy ra những thông số cần thiết của dữ liệu(đã qua xử lý) để tạo ma trận thưa","metadata":{}},{"cell_type":"code","source":"##Tạo một mảng dữ liệu 3 giá trị như sau: Hàng(dòng dữ liệu thứ mấy),cột(từ đó là từ thứ bao nhiêu trong vocabulary),giá trị(từ đó xuất hiện bao nhiêu lần)\ndef createConfigValueForMatixSprase(doc,vocab):\n    rows = []\n    columns = []\n    values = []\n    \n    indexSentence = 0; # index của dữ liệu câu hỏi\n    for sentence in tqdm(doc): #lặp qua từng câu hỏi trong kho dữ liệu\n        dictOfWord = {}  #Khởi tạo tập hợp các từ có trong câu\n        for word in sentence.split(): #Xét các từ bên trong một câu\n            try: # Nếu tập hợp đã có từ đang xét,số lượng từ đó trong dict tăng lên 1\n                dictOfWord[word] += 1;\n            except: # Nếu tập hợp chưa có từ đang xét,thêm từ đó vào trong dict,với số lượng từ bằng 1\n                dictOfWord.update({word:1})\n        #Lặp qua tất cả các từ có trong từ điển của câu.\n        for word in dictOfWord.keys():\n            rows.append(indexSentence) #Thêm dữ liệu thứ tự dòng\n            try:\n                columns.append(vocab[word]) #Thêm dữ liệu thứ thự cột(Vị trí của từ trong index_Vocabulary)\n            except:\n                columns.append(len(vocab)) # Nếu từ đó không có trong vocabulary,cho nó xuống hang cuối cùng của từ vựng\n            values.append(dictOfWord[word]) # Thêm dữ liệu về số lần xuất hiện của từ đó\n        indexSentence += 1;\n    return (rows,columns,values)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:21.051522Z","iopub.execute_input":"2021-05-25T03:42:21.051847Z","iopub.status.idle":"2021-05-25T03:42:21.063592Z","shell.execute_reply.started":"2021-05-25T03:42:21.051817Z","shell.execute_reply":"2021-05-25T03:42:21.062404Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Hàm tạo ma trận thưa","metadata":{}},{"cell_type":"code","source":"from scipy.sparse import coo_matrix\ndef createSparseMatrix(data,vocab):\n    rows,columns,values = createConfigValueForMatixSprase(data,vocab)\n    return coo_matrix((values,(rows,columns)),shape=(len(data),len(vocab) + 1))","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:21.065045Z","iopub.execute_input":"2021-05-25T03:42:21.065501Z","iopub.status.idle":"2021-05-25T03:42:21.078004Z","shell.execute_reply.started":"2021-05-25T03:42:21.065467Z","shell.execute_reply":"2021-05-25T03:42:21.076990Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Khởi tạo ma trận thưa cho tập traning ","metadata":{}},{"cell_type":"code","source":"train_simple_X = createSparseMatrix(train_processed_doc,index_Vocabulary)\nlabel_X = train.target","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:21.079402Z","iopub.execute_input":"2021-05-25T03:42:21.079781Z","iopub.status.idle":"2021-05-25T03:42:26.935391Z","shell.execute_reply.started":"2021-05-25T03:42:21.079722Z","shell.execute_reply":"2021-05-25T03:42:26.934396Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    ","metadata":{}},{"cell_type":"markdown","source":"    Khởi tạo ma trận thưa cho tập validation","metadata":{}},{"cell_type":"code","source":"valid_simple_X = createSparseMatrix(val_processed_doc,index_Vocabulary)\nlabel_y = val.target","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:26.936609Z","iopub.execute_input":"2021-05-25T03:42:26.936897Z","iopub.status.idle":"2021-05-25T03:42:29.430181Z","shell.execute_reply.started":"2021-05-25T03:42:26.936870Z","shell.execute_reply":"2021-05-25T03:42:29.428764Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **Nhận xét** :\n> > * Khi lần đầu thử ,xử lý dữ liệu ở tập valid bị lỗi do có một số từ xuất hiện ở tập train nhưng lại không xuất hiện ở tập valid.Nên tôi sẽ quy ước những từ không gặp ở tập train,nhưng gặp ở tập valid là những từ ở cột cuối cùng\n> > * Cách sử dụng ma trận rỗng có tốc độ nhanh hơn và ổn hơn rất nhiều so với cách dùng ma trận đặc ","metadata":{}},{"cell_type":"markdown","source":"># **1.3-Kết quả** : \n>  * **Lấy ra từ vựng** :Việc lấy ra từ vựng khá nhanh chóng và chuẩn xác\n>  * **Kiến tạo vector feature**:\n>   * Khi sử dụng phương án 1:thời gian để tạo một ma trận dữ liệu cho toàn bộ dữ liệu cực kỳ lâu do số lượng từ mỗi câu ít hơn rất nhiều số lượng từ của cả đoạn văn,không có khả năng ứng dụng trong việc xử lý tạo vector feature.\n>   * Khi sử dụng phương án 2:thời gian để tạo một ma trận thưa cho toàn bộ dữ liệu nhanh hơn rất nhiều so với phương án một,có khả năng áp dụng trong việc xử lý dữ liệu.","metadata":{}},{"cell_type":"markdown","source":"> # **1.4-Bình luận** : \n>  * **Vấn đề nảy sinh**:\n>   * Việc xử lý thủ công,không sử dụng thư viện như thế này rất tốn thời gian và công sức,đòi hỏi phải viết code xử lý khá nhiều và mất thời gian.\n>  * **Ý kiến**:Giải quyết như thế này chưa ổn lắm\n>   * Ưu điểm:Đơn giản,dễ hiểu,dễ áp dụng cho cả người thực hiện lần người đọc;người viết code có thể quản lý và kiểm soát hoàn toàn bag-of-word\n>   * Nhược điểm:rất rõ ràng,do sử dụng bag-of-word nên thông tin về thứ tự từ trong câu đã bị mất.Hơn nữa,việc trích chọn các từ đặc trưng hơn các từ khác là vô cùng khó khăn vì không thể biết được độ đặc trưng của một từ thông qua số lượng từ đó được.Số lượng feature = từ vựng nên rất tốn bộ nhớ mà mất thời gian trong việc traing các model\n","metadata":{}},{"cell_type":"markdown","source":"> #  **2.Tạo Vector từ dữ liệu bằng tf-idf**","metadata":{}},{"cell_type":"markdown","source":"> # **2.1-Vấn đề**:\nViệc xử lý dữ liệu theo cách thủ công rõ ràng đã để lộ điểm yếu về sự thiếu hụt các trọng số thể hiện đặc trưng cho các từ vựng cũng như lượng bộ nhớ quá lớn cần sử dụng.Vì thế cần sử dụng thư viện tf-idf khi xử lý để có trọng số tf-idf(trọng số đặc trưng) cho từ word,nhằm mục đích chọn lựa những từ đặc trưng nhất để training model hiệu quả hơn,giảm thời gian và bộ nhớ xuống.Vấn đề đặt ra là dùng bao nhiêu từ thì hợp lý,và chọn những từ nào làm feature để lượng thông tin bị mất đi ít nhất.","metadata":{}},{"cell_type":"markdown","source":"># **2.2-Giải pháp**:\n> * **Lấy ra từ vựng**:Chúng ta cần thử nghiệm,điều chỉnh các tham số để có được một bộ tham số tốt nhất,sau đó sử dụng các tham số đó.","metadata":{}},{"cell_type":"markdown","source":"> ># 2.2.1-Lấy ra từ vựng:","metadata":{}},{"cell_type":"code","source":"from sklearn.feature_extraction.text import TfidfVectorizer\ntf = TfidfVectorizer(analyzer='word',stop_words='english',min_df=4,max_df=0.8) #khởi tạo,cài đặt số lượng tối đa cho feature\ntf.fit(train_processed_doc) #các feature(từ vựng) sẽ được lấy từ tập traning","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:29.431799Z","iopub.execute_input":"2021-05-25T03:42:29.432080Z","iopub.status.idle":"2021-05-25T03:42:33.572358Z","shell.execute_reply.started":"2021-05-25T03:42:29.432054Z","shell.execute_reply":"2021-05-25T03:42:33.571319Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **Nhận xét**:Sau nhiều lần thử nghiệm,bộ tham số trên(analyzer='word',stop_words='english',min_df=4,max_df=0.8) có thể cho kết quả khá tốt","metadata":{}},{"cell_type":"markdown","source":"    Lấy tên các feature sau khi tf-idf đã học xong từ vựng.","metadata":{}},{"cell_type":"code","source":"feature_names = tf.get_feature_names()","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:33.573616Z","iopub.execute_input":"2021-05-25T03:42:33.574177Z","iopub.status.idle":"2021-05-25T03:42:33.596296Z","shell.execute_reply.started":"2021-05-25T03:42:33.574135Z","shell.execute_reply":"2021-05-25T03:42:33.595047Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> ># 2.2.2-Tạo Ma trận thưa bằng TfidfVectorizer ","metadata":{}},{"cell_type":"code","source":"train_tfidf_X = tf.transform(train_processed_doc)\nval_tfidf_X = tf.transform(val_processed_doc)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:33.598057Z","iopub.execute_input":"2021-05-25T03:42:33.598544Z","iopub.status.idle":"2021-05-25T03:42:39.242634Z","shell.execute_reply.started":"2021-05-25T03:42:33.598491Z","shell.execute_reply":"2021-05-25T03:42:39.241538Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(len(train_processed_doc))\nprint(len(feature_names))","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:39.243931Z","iopub.execute_input":"2021-05-25T03:42:39.244209Z","iopub.status.idle":"2021-05-25T03:42:39.249241Z","shell.execute_reply.started":"2021-05-25T03:42:39.244182Z","shell.execute_reply":"2021-05-25T03:42:39.248246Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"># **2.3-Kết quả**:\n> * **Lấy ra từ vựng**:TfidfVectorizer có các hàm phụ trợ lấy từ vựng rất tốt nhờ các tham số như 'min_df' và 'max_df'\n> * Từ vựng lấy ra thể hiện tính đăch trưng cao(Sẽ được giải thích ở phần kết luận ở cuối bài)","metadata":{}},{"cell_type":"markdown","source":"> # **2.4-Bình luận** : \n>  * **Ý kiến**:\n>   * Ưu điểm:Dễ áp dụng,không cần code nhiều,nhanh chóng,tiện lợi,tiết kiệm thời gian\n>   * Nhược điểm:Mất khá nhiều thời gian để hiểu sâu về tham số và cách hoạt động của tfidfVectorizer.","metadata":{}},{"cell_type":"markdown","source":"# **Chương 3:Training Model và thử nghiệm**","metadata":{}},{"cell_type":"markdown","source":"    Import các thư viện cần thiết","metadata":{}},{"cell_type":"code","source":"##Import\nfrom sklearn.naive_bayes import MultinomialNB, BernoulliNB\nfrom sklearn.linear_model import Perceptron\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.ensemble import RandomForestClassifier\nfrom sklearn.ensemble import GradientBoostingClassifier\nfrom sklearn.ensemble import AdaBoostClassifier\nfrom sklearn.decomposition import TruncatedSVD\nfrom sklearn.metrics import classification_report\nfrom sklearn import svm","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:39.250615Z","iopub.execute_input":"2021-05-25T03:42:39.250907Z","iopub.status.idle":"2021-05-25T03:42:39.480619Z","shell.execute_reply.started":"2021-05-25T03:42:39.250879Z","shell.execute_reply":"2021-05-25T03:42:39.479656Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> #  **1.Hàm đánh giá**\n>   **Chúng ta sẽ dựa vào 4 chỉ số Precision,Recall,F1-Score và Accurancy để đánh giá tính hiệu quả của model**","metadata":{}},{"cell_type":"code","source":"#Đánh giá model qua tập validation\ndef validModel(Model,valid_model_X):\n    valid_y = val.target;\n    pred_y = Model.predict(valid_model_X)\n    print(classification_report(label_y,pred_y))","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:39.482313Z","iopub.execute_input":"2021-05-25T03:42:39.482753Z","iopub.status.idle":"2021-05-25T03:42:39.489692Z","shell.execute_reply.started":"2021-05-25T03:42:39.482708Z","shell.execute_reply":"2021-05-25T03:42:39.488529Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> #  **2.Sử dụng LogisticRegression**","metadata":{}},{"cell_type":"markdown","source":"> # **2.1-Với dữ liệu được xử lý bằng cách đơn giản**","metadata":{}},{"cell_type":"markdown","source":"> **Training**","metadata":{}},{"cell_type":"code","source":"simple_logistic = LogisticRegression(penalty=\"l2\", C=1,max_iter=800) \nsimple_logistic.fit(train_simple_X,label_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:42:39.490775Z","iopub.execute_input":"2021-05-25T03:42:39.491102Z","iopub.status.idle":"2021-05-25T03:43:23.214478Z","shell.execute_reply.started":"2021-05-25T03:42:39.491012Z","shell.execute_reply":"2021-05-25T03:43:23.213451Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Đánh giá**","metadata":{}},{"cell_type":"code","source":"validModel(simple_logistic,valid_simple_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:43:23.219478Z","iopub.execute_input":"2021-05-25T03:43:23.222129Z","iopub.status.idle":"2021-05-25T03:43:23.455586Z","shell.execute_reply.started":"2021-05-25T03:43:23.222070Z","shell.execute_reply":"2021-05-25T03:43:23.454564Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> # **2.2-Với dữ liệu được xử lý bằng tf-idf**","metadata":{}},{"cell_type":"markdown","source":"> **Training**","metadata":{}},{"cell_type":"code","source":"tfidf_logistic = LogisticRegression(penalty=\"l2\", C=1,max_iter=400) \ntfidf_logistic.fit(train_tfidf_X,label_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:43:23.457042Z","iopub.execute_input":"2021-05-25T03:43:23.457466Z","iopub.status.idle":"2021-05-25T03:43:38.641362Z","shell.execute_reply.started":"2021-05-25T03:43:23.457429Z","shell.execute_reply":"2021-05-25T03:43:38.640317Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Đánh giá**","metadata":{}},{"cell_type":"code","source":"validModel(tfidf_logistic,val_tfidf_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:43:38.646404Z","iopub.execute_input":"2021-05-25T03:43:38.649114Z","iopub.status.idle":"2021-05-25T03:43:38.864309Z","shell.execute_reply.started":"2021-05-25T03:43:38.649054Z","shell.execute_reply":"2021-05-25T03:43:38.862219Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> > **Nhận xét**: Các chỉ số khá tốt,f1-score = 0.76","metadata":{}},{"cell_type":"markdown","source":"> #  **3.Sử dụng MultinomialNB**","metadata":{}},{"cell_type":"markdown","source":"> # **3.1-Với dữ liệu được xử lý bằng cách đơn giản**","metadata":{}},{"cell_type":"markdown","source":"    Training Model với dữ liệu được xử lý theo cách đơn giản","metadata":{}},{"cell_type":"code","source":"#Training Model MultinomialNB()\nsimple_multi_NB_model = MultinomialNB()\nsimple_multi_NB_model.fit(train_simple_X,label_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:43:38.865881Z","iopub.execute_input":"2021-05-25T03:43:38.866328Z","iopub.status.idle":"2021-05-25T03:43:39.017785Z","shell.execute_reply.started":"2021-05-25T03:43:38.866295Z","shell.execute_reply":"2021-05-25T03:43:39.016793Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Đánh giá**","metadata":{}},{"cell_type":"code","source":"validModel(simple_multi_NB_model,valid_simple_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:43:39.021081Z","iopub.execute_input":"2021-05-25T03:43:39.021392Z","iopub.status.idle":"2021-05-25T03:43:39.221928Z","shell.execute_reply.started":"2021-05-25T03:43:39.021361Z","shell.execute_reply":"2021-05-25T03:43:39.220925Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    92.68 %,khá ổn","metadata":{}},{"cell_type":"markdown","source":"> # **3.2-Với dữ liệu được xử lý bằng tf-idf**","metadata":{}},{"cell_type":"markdown","source":"    Training Model với dữ liệu được xử lý theo tfidf","metadata":{}},{"cell_type":"code","source":"tfidf_multi_NB_model = MultinomialNB()\ntfidf_multi_NB_model.fit(train_tfidf_X,label_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:43:39.223139Z","iopub.execute_input":"2021-05-25T03:43:39.223443Z","iopub.status.idle":"2021-05-25T03:43:39.271946Z","shell.execute_reply.started":"2021-05-25T03:43:39.223395Z","shell.execute_reply":"2021-05-25T03:43:39.270889Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Đánh giá**","metadata":{}},{"cell_type":"code","source":"validModel(tfidf_multi_NB_model,val_tfidf_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:43:39.273411Z","iopub.execute_input":"2021-05-25T03:43:39.273820Z","iopub.status.idle":"2021-05-25T03:43:39.426367Z","shell.execute_reply.started":"2021-05-25T03:43:39.273777Z","shell.execute_reply":"2021-05-25T03:43:39.425122Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> # **4.Sử dụng BernoulliNB**","metadata":{}},{"cell_type":"markdown","source":"> # **4.1-Với dữ liệu được xử lý bằng cách đơn giản**","metadata":{}},{"cell_type":"markdown","source":"    Training Mode với dữ liệu được xử lý bằng cách đơn giản","metadata":{}},{"cell_type":"code","source":"simple_ber_NB_model = BernoulliNB()\nsimple_ber_NB_model.fit(train_simple_X,label_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:43:39.427624Z","iopub.execute_input":"2021-05-25T03:43:39.427925Z","iopub.status.idle":"2021-05-25T03:43:39.596664Z","shell.execute_reply.started":"2021-05-25T03:43:39.427897Z","shell.execute_reply":"2021-05-25T03:43:39.595525Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Đánh giá**","metadata":{}},{"cell_type":"code","source":"validModel(simple_ber_NB_model,valid_simple_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:43:39.597812Z","iopub.execute_input":"2021-05-25T03:43:39.598075Z","iopub.status.idle":"2021-05-25T03:43:39.809279Z","shell.execute_reply.started":"2021-05-25T03:43:39.598050Z","shell.execute_reply":"2021-05-25T03:43:39.808261Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    93,09% ,cao hơn MultinomialNB một chút","metadata":{}},{"cell_type":"markdown","source":"> # **4.2-Với dữ liệu được xử lý bằng tf-idf**","metadata":{}},{"cell_type":"markdown","source":"> **Training**","metadata":{}},{"cell_type":"code","source":"tfidf_ber_NB_model = BernoulliNB()\ntfidf_ber_NB_model.fit(train_tfidf_X,label_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:43:39.812593Z","iopub.execute_input":"2021-05-25T03:43:39.812919Z","iopub.status.idle":"2021-05-25T03:43:39.874124Z","shell.execute_reply.started":"2021-05-25T03:43:39.812888Z","shell.execute_reply":"2021-05-25T03:43:39.873138Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Đánh giá**","metadata":{}},{"cell_type":"code","source":"validModel(tfidf_ber_NB_model,val_tfidf_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:43:39.875590Z","iopub.execute_input":"2021-05-25T03:43:39.876017Z","iopub.status.idle":"2021-05-25T03:43:40.035918Z","shell.execute_reply.started":"2021-05-25T03:43:39.875972Z","shell.execute_reply":"2021-05-25T03:43:40.034803Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> # **5.Sử dụng XGBoost**","metadata":{}},{"cell_type":"markdown","source":"> # **5.1-Với dữ liệu được xử lý bằng cách đơn giản**","metadata":{}},{"cell_type":"markdown","source":"> **Training**","metadata":{}},{"cell_type":"code","source":"simple_GBClassifier_model = GradientBoostingClassifier(max_features=0.2,n_estimators=150)\nsimple_GBClassifier_model.fit(train_simple_X,label_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:43:40.037374Z","iopub.execute_input":"2021-05-25T03:43:40.037794Z","iopub.status.idle":"2021-05-25T03:44:04.238013Z","shell.execute_reply.started":"2021-05-25T03:43:40.037749Z","shell.execute_reply":"2021-05-25T03:44:04.236326Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Đánh giá**","metadata":{}},{"cell_type":"code","source":"validModel(simple_GBClassifier_model,valid_simple_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:44:04.238889Z","iopub.status.idle":"2021-05-25T03:44:04.239306Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> # **5.2-Với dữ liệu được xử lý bằng tf-idf**","metadata":{}},{"cell_type":"markdown","source":"> **Training**","metadata":{}},{"cell_type":"code","source":"tfidf_GBClassifier_model = GradientBoostingClassifier(max_features=0.2, n_estimators=150)\ntfidf_GBClassifier_model.fit(train_tfidf_X,label_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:44:04.240501Z","iopub.status.idle":"2021-05-25T03:44:04.240923Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Đánh giá**","metadata":{}},{"cell_type":"code","source":"validModel(tfidf_GBClassifier_model,val_tfidf_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:44:04.241871Z","iopub.status.idle":"2021-05-25T03:44:04.242282Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> # **6.Sử dụng AdaBoostClassifier**","metadata":{}},{"cell_type":"markdown","source":"> # **6.1-Với dữ liệu được xử lý bằng tf-idf**","metadata":{}},{"cell_type":"markdown","source":"> **Training**","metadata":{}},{"cell_type":"code","source":"tfidf_AdaBoost_model = AdaBoostClassifier(n_estimators=1000)\ntfidf_AdaBoost_model.fit(train_tfidf_X,label_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:44:04.243069Z","iopub.status.idle":"2021-05-25T03:44:04.243460Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Đánh giá**","metadata":{}},{"cell_type":"code","source":"validModel(tfidf_AdaBoost_model,val_tfidf_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:44:04.244538Z","iopub.status.idle":"2021-05-25T03:44:04.244954Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# **Chương 4:Tổng kết**\n* **Precision**:Có bao nhiêu % mẫu positive được mô hình dự đoán chính xác so với tổng số mẫu positive được mô hình dự đoán.\n* **Recall**: Có bao nhiêu % mẫu positive được mô hình dự đoán chính xác so với tổng số mẫu positive trong thực tế.\n* **F1-Score**: Một trị số trung bình giữa Precision và Recall,trị số này càng cao thì mô hình càng tốt\n* **Accurancy**: Có bao nhiêu % mẫu được dự đoán chính xác so với tổng toàn bộ số mẫu đang có.","metadata":{}},{"cell_type":"markdown","source":"> # **1.Xử lý dữ liệu**","metadata":{}},{"cell_type":"markdown","source":"> # **1.1-So sánh các chỉ số Precision,Recall và F1-Score,Accurancy trên một số mô hình**","metadata":{}},{"cell_type":"markdown","source":"> > # 1.1.1-Biểu đồ so sánh","metadata":{}},{"cell_type":"markdown","source":"> > **Biểu đồ 3 chỉ số Precision,Recall và F1-Score**","metadata":{}},{"cell_type":"markdown","source":"![image.png](attachment:5c8971e8-263c-4401-8af1-596b2d7d0b12.png)","metadata":{},"attachments":{"5c8971e8-263c-4401-8af1-596b2d7d0b12.png":{"image/png":"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"}}},{"cell_type":"markdown","source":"> > **Biểu đồ chỉ số accurancy**","metadata":{}},{"cell_type":"markdown","source":"> > # 1.1.2-Nhận xét\n> > * **Có thể thấy rằng,các chỉ số Precision,Recall,F1-Score không khác nhau là mấy khi dự đoán những câu hỏi không toxic ở cả hai trường hợp,nhưng lại khác biệt rất rõ rệt khi dự đoán các câu hỏi toxic.Recall và F1-Score được tăng lên rõ rệt nhờ resample**\n> > * **Mặc dù accurancy khi không dùng resample cao hơn khi dùng resample(Vì chúng ta sử dụng under resampling),nhưng F1-Score lại thấp hơn khá nhiều.Rõ ràng,khi không sử dụng resample,số label toxic ít hơn số label không toxic quá nhiều(khoảng 16 lần) dẫn đến accurancy không còn đáng tin cậy,nên có thể nhận xét rằng,logistic regression tốt hơn khi sử dụng resample**","metadata":{}},{"cell_type":"markdown","source":"> # **1.2-So sánh các chỉ số Precision,Recall và F1-Score trên mô hình XGBoost(Sử dụng tfidf)**","metadata":{}},{"cell_type":"markdown","source":"> > **Biểu đồ 3 chỉ số Precision,Recall và F1-Score**","metadata":{}},{"cell_type":"markdown","source":"![image.png](attachment:2ffccbb7-c31b-4217-8fe0-2c5f16db4e01.png)","metadata":{},"attachments":{"2ffccbb7-c31b-4217-8fe0-2c5f16db4e01.png":{"image/png":"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"}}},{"cell_type":"markdown","source":"> > **Biểu đồ chỉ số accurancy**","metadata":{}},{"cell_type":"markdown","source":"> > # 1.2.2-Nhận xét\n> > * **Ở các biểu đồ phần 1.2 này,chúng ta còn thấy rõ ràng hơn việc thiên lệch label quá lớn đã ảnh hưởng đến model nhiều như thế nào.Việc XGBoost cố gắng fit với lượng dữ liệu dán nhãn label không toxic(nhiều gấp 16 lần nhãn không toxic) đã dẫn đến chí số Recall và F1-Score của các câu hỏi toxic có giá trị rất tồi.Ngay cả khi sử dụng resample(nhãn không toxic nhiều gấp 3 lần nhãn toxic),2 chỉ số này ở các câu hỏi toxic vẫn thấp hơn nhiều ở câu hỏi không toxic.**\n> > * **Mặc dù accurancy khi không sử dụng resample là rất tốt,nhưng sự thiên lệch dữ liệu khiến cho accurancy không còn đáng tin cậy.Khi nhìn vào chỉ số Recall và F1-Score(rất tồi) khi không sử dụng resample,chúng ta có thể thấy rõ được sự không đáng tin cậy đó.**\n> > * **Thời gian khi dùng resample cũng nhanh hơn nhiều so với lúc không dùng resample**","metadata":{}},{"cell_type":"markdown","source":"> # **1.2-Bình luận**\n> * **Kết quả so sánh trên đã cho thấy sự ảnh hưởng tiêu cực của việc thiên lệch nhãn tới các mô hình(Nhất là những mô hình học kết hợp như XGBoost) và cho thấy tính hiệu quả của resample với những mẫu dữ liệu bị thiên lệch.**\n> * **Chúng ta sẽ sử dụng các mẫu dữ liệu sử resample trong việc training model để tăng hiệu quả của model,giảm ảnh hương tiêu cực của sự thiên lệch model**\n> * **Sử dụng thêm tfidf để giảm lượng từ vựng(Giảm feature mà vẫn đảm bảo tính đặc trưng của câu) để giảm thời gian training của các mô hình học kết hợp,tránh việc bị overfitting**","metadata":{}},{"cell_type":"markdown","source":"> # **2.Đánh giá về các loại model**","metadata":{}},{"cell_type":"markdown","source":"> # **2.1-Chỉ số đánh giá sự hiệu quả của các model(xử lý dữ liệu bằng resample,vector được tạo bởi tf-idf)**","metadata":{}},{"cell_type":"markdown","source":"> > # 2.1.1-So sánh các chỉ số Precision,Recall và F1-Score giữa các model(Bên phải là toxic,bên trái là không toxic tính từ tên của model)","metadata":{}},{"cell_type":"markdown","source":"> > > **Biểu đồ**","metadata":{}},{"cell_type":"markdown","source":"![image.png](attachment:d769bc94-6f63-457b-9f23-e95f14268f1e.png)","metadata":{},"attachments":{"d769bc94-6f63-457b-9f23-e95f14268f1e.png":{"image/png":"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"}}},{"cell_type":"markdown","source":"> > > **Bình luận**\n> > > * **Một số quan sát thú vị**:\n> > >  * Trong tất cả các model trên,chỉ số recall luôn nhỏ hơn chỉ số precision ở các câu hỏi toxic,ngoại trừ ở mô hình BernoulliNB(Recall là tỉ lệ bao nhiêu mẫu thực tế được xác định đúng).\n> > >  * XGBoost có vẻ như không hiệu quả khi giải bài toán xử lý dữ liệu text(gồm rất nhiều feature) như thế này khi mà chỉ số Recall và F1 thấp hơn nhiều so với các model khác ở model toxic.\n> > >  * Mặc dù adaboost có chỉ số F1-Score thấp hơn một chút so với Logistic Regression và  BernoulliNB(ở câu hỏi toxic),nhưng nó lại có thể tiếp tục cải thiện F1-Score bằng cách tăng n_estimators lên.","metadata":{}},{"cell_type":"markdown","source":"> > # 3.1.2-So sánh chỉ số Accurancy,và F1-Score(Toxic) giữa các model","metadata":{}},{"cell_type":"markdown","source":"> > > **Biểu đồ**","metadata":{}},{"cell_type":"markdown","source":"![image.png](attachment:22f4bdb3-d3fb-4d5e-a06c-d2162a121e71.png)","metadata":{},"attachments":{"22f4bdb3-d3fb-4d5e-a06c-d2162a121e71.png":{"image/png":"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"}}},{"cell_type":"markdown","source":"> > > **Nhận xét**:\n> > > * Các mô hình truyền thống tỏ ra rất hiệu quả với bài toán phân loại dữ liệu dạng text(Logistic Regression,Naive Bayes Classifier).\n> > > * Dù XGBoost chỉ số Accurancy khá cao,nhưng chỉ số F1-Score cho các câu hỏi Toxic(Câu hỏi cần mô hình xác định) lại khá thấp,cho thấy sự kém hiệu quả của XGBoost với dạng bài toán phân loại text này so với các mô hình truyền thống khác.","metadata":{}},{"cell_type":"markdown","source":"> # **2.2-Bình luận**:\n> * **Cả về chỉ số F1-Score lẫn Accurancy,các mô hình phân loại truyền thống đang tỏ ra hiệu quả hơn các mô hình học kết hợp**\n> * **Với mỗi loại mô hình(truyền thống và esemble) tôi sẽ chọn một đại diện để thử nghiệm việc test**\n>  * **Mô hình truyền thống**:\n>   * **MultinomialNB** có chỉ số Recall,F1-Score và Accurancy của nó là tốt nhất.Chúng ta cần tỉ lệ dự đoán chính xác trên tổng mẫu thực tế của hơn là trên tổng mẫu dự đoán của mô hình(mẫu toxic).\n>   * **Logistic Regression** có các chỉ số F1-Score và Accurancy rất tốt,đặc biệt là khi sử dụng toàn bộ từ vựng thì F1-Score của mô hình này lên tới 0.8(đối với câu hỏi Toxic),cao nhất trong tất cả các mô hình.\n>  * **Mô hình Esemble**:Có các chỉ số khá hiệu quả và còn có thể cải thiện tính hiệu quả.","metadata":{}},{"cell_type":"markdown","source":"> # **4.Training model với toàn bộ dữ liệu train và nộp bài**","metadata":{}},{"cell_type":"markdown","source":"> **Khởi tạo tfidf**","metadata":{}},{"cell_type":"code","source":"all_tf = TfidfVectorizer(analyzer='word',stop_words='english',min_df=4,max_df=0.8)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:44:11.956264Z","iopub.execute_input":"2021-05-25T03:44:11.956761Z","iopub.status.idle":"2021-05-25T03:44:11.961871Z","shell.execute_reply.started":"2021-05-25T03:44:11.956701Z","shell.execute_reply":"2021-05-25T03:44:11.960518Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Chọn mô hình LogisticRegression(+Simple process data)**","metadata":{}},{"cell_type":"code","source":"model = LogisticRegression(penalty=\"l2\", C=1,max_iter=800) \n#model = AdaBoostClassifier(n_estimators=2000)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:44:13.808595Z","iopub.execute_input":"2021-05-25T03:44:13.808968Z","iopub.status.idle":"2021-05-25T03:44:13.813083Z","shell.execute_reply.started":"2021-05-25T03:44:13.808938Z","shell.execute_reply":"2021-05-25T03:44:13.812256Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Chia lại dữ liệu để cân bằng các label**","metadata":{}},{"cell_type":"code","source":"#from sklearn.utils import resample\nall_sincere = train_data[train_data.target == 0]\nall_insincere = train_data[train_data.target == 1]\ntrain_data = pd.concat([resample(all_sincere,replace = True,n_samples = len(all_insincere)*3), all_insincere])\nprint(len(train_data[train_data.target == 0]))\nprint(len(train_data[train_data.target == 1]))","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:44:15.378910Z","iopub.execute_input":"2021-05-25T03:44:15.379235Z","iopub.status.idle":"2021-05-25T03:44:15.498016Z","shell.execute_reply.started":"2021-05-25T03:44:15.379208Z","shell.execute_reply":"2021-05-25T03:44:15.496856Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Loại bỏ các nhiễu khỏi dữ liệu traning**","metadata":{}},{"cell_type":"code","source":"all_train_processed_data = Preprocess(train_data.question_text)\nall_label_X = raw_train_data.target","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:44:16.867616Z","iopub.execute_input":"2021-05-25T03:44:16.867963Z","iopub.status.idle":"2021-05-25T03:44:29.541367Z","shell.execute_reply.started":"2021-05-25T03:44:16.867933Z","shell.execute_reply":"2021-05-25T03:44:29.540627Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Vector hóa dữ liệu Traning**","metadata":{}},{"cell_type":"markdown","source":"    Tfidf","metadata":{}},{"cell_type":"code","source":"#all_tf.fit(all_train_processed_data)\n#all_tfidf_train_X = all_tf.transform(all_train_processed_data)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:44:04.250984Z","iopub.status.idle":"2021-05-25T03:44:04.251409Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Simple","metadata":{}},{"cell_type":"code","source":"all_vocabulary = get_vocab(all_train_processed_data) # Lấy vocabulary từ dữ liệu training\nindex_all_vocabulary = vocabulary.copy() # Khởi tạo directory chứa tên của từ và index của từ đó\ninitIndexVocabulary(index_all_vocabulary)\nall_simple_train_X = createSparseMatrix(all_train_processed_data,index_all_vocabulary)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:44:29.542634Z","iopub.execute_input":"2021-05-25T03:44:29.543093Z","iopub.status.idle":"2021-05-25T03:44:39.561156Z","shell.execute_reply.started":"2021-05-25T03:44:29.543053Z","shell.execute_reply":"2021-05-25T03:44:39.560059Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Training mô hình.**","metadata":{}},{"cell_type":"code","source":"model.fit(all_simple_train_X,all_label_X)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:48:22.245031Z","iopub.execute_input":"2021-05-25T03:48:22.245399Z","iopub.status.idle":"2021-05-25T03:49:26.369193Z","shell.execute_reply.started":"2021-05-25T03:48:22.245369Z","shell.execute_reply":"2021-05-25T03:49:26.368012Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Vector hóa dữ liệu test**","metadata":{}},{"cell_type":"markdown","source":"    Simple","metadata":{}},{"cell_type":"code","source":"all_test_processed_data = Preprocess(raw_test_data.question_text)\ntestt_padded = createSparseMatrix(all_test_processed_data,index_all_vocabulary)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:49:26.370880Z","iopub.execute_input":"2021-05-25T03:49:26.371275Z","iopub.status.idle":"2021-05-25T03:49:49.569395Z","shell.execute_reply.started":"2021-05-25T03:49:26.371233Z","shell.execute_reply":"2021-05-25T03:49:49.568350Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"    Tfidf","metadata":{}},{"cell_type":"code","source":"#all_test_processed_data = Preprocess(raw_test_data.question_text)\n#testt_padded = all_tf.transform(all_test_processed_data)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:44:04.256532Z","iopub.status.idle":"2021-05-25T03:44:04.256976Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":">  **Predict dữ liệu**","metadata":{}},{"cell_type":"code","source":"y_test_pre = model.predict(testt_padded)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:49:49.571250Z","iopub.execute_input":"2021-05-25T03:49:49.571560Z","iopub.status.idle":"2021-05-25T03:49:49.733470Z","shell.execute_reply.started":"2021-05-25T03:49:49.571528Z","shell.execute_reply":"2021-05-25T03:49:49.732576Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> **Tạo bài submit cho bộ dữ liệu test**","metadata":{}},{"cell_type":"code","source":"submit=pd.DataFrame()\nsubmit[\"qid\"]=raw_test_data.qid\nsubmit[\"prediction\"]=y_test_pre\nsubmit.to_csv(\"submission.csv\",index=False)","metadata":{"execution":{"iopub.status.busy":"2021-05-25T03:49:49.734877Z","iopub.execute_input":"2021-05-25T03:49:49.735160Z","iopub.status.idle":"2021-05-25T03:49:50.644761Z","shell.execute_reply.started":"2021-05-25T03:49:49.735133Z","shell.execute_reply":"2021-05-25T03:49:50.643765Z"},"trusted":true},"execution_count":null,"outputs":[]}]}