31. Linear Regression with TF(sol).ipynb

{"nbformat":4,"nbformat_minor":0,"metadata":{"colab":{"name":"31. Linear Regression with TF(sol).ipynb","version":"0.3.2","views":{},"default_view":{},"provenance":[],"collapsed_sections":[]},"kernelspec":{"display_name":"Python 3","language":"python","name":"python3"}},"cells":[{"metadata":{"id":"C_A27R35DJnv","colab_type":"text"},"cell_type":"markdown","source":["# 01. Linear Regression with TF\n","\n","<p style=\"text-align: right;\">\n","rayleigh@dplus.company<br>\n","Your name :\n","</p>\n","\n","### 텐서플로 실습시간 그 두번째!<br>\n","\n","#### 선행지식\n","1. 모델링을 한다는 것에 대한 이해.\n","2. Supervised Learning 중 Regression에 대한 기본 지식.\n","3. 리니어 리그레션에 대한 기본 구조와 개념\n","4. 앞서 진행한 실습에 대한 추억(00. TensorFlow Basics, 01. Linear Regression with TF)\n","\n","#### 실습목표\n","1. Linear Regression의 구조를 Graph로 그려낼 수 있다.\n","2. tf.Variable과 초기화, Optimizer를 사용할 수 있다.\n","3. 그려낸 Graph를 텐서플로우를 이용해 코딩할 수 있다.\n","4. (Optional) Statistical Model의 Signal & Ratio 를 이해한다.\n","5. (Optional) Encoding이 무엇인지 이해한다. --> Dummy Variable, Onehot-encoding 등의 필요성을 이해한다.\n","\n","\n","#### 사용데이터.\n","\n","01. Univariate Linear Regression : Lemonade.csv (가상 데이터셋)\n","02. Multivariate Linear Regression : http://archive.ics.uci.edu/ml/datasets/Bike+Sharing+Dataset"]},{"metadata":{"id":"8ja0novwDJnx","colab_type":"text"},"cell_type":"markdown","source":["# Univariate Linear Regression"]},{"metadata":{"id":"hekahF2HDJny","colab_type":"text"},"cell_type":"markdown","source":["## 00. 라이브러리 불러오기"]},{"metadata":{"id":"majv4I0ODJnz","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["%matplotlib inline\n","%config InlineBackend.figure_format = 'retina'\n","\n","import pandas as pd\n","import numpy as np\n","import tensorflow as tf\n","import matplotlib.pyplot as plt\n","import time"],"execution_count":0,"outputs":[]},{"metadata":{"id":"Wq41A_oRDJn3","colab_type":"text"},"cell_type":"markdown","source":["## 01. 데이터 불러오기 & 문제정의 & 전처리\n","\n","여러분은 이동식 레모네이드 카페, '롤로노아'의 사장입니다.<br>\n","차별화된 맛과, 원재료의 신선함. 한 번도 안먹어본 사람은 있을 수 있어도, 한 번만 마셔본 사람은 없다!<br>\n","수년간 늦봄~초가을에 포장마차를 끌고 행복을 팔며 느꼈던 것은...<br>\n","#### 여러분의 불행은 팔지못하고 남은 재료의 양과 비례한다는 것입니다.<br>\n","\n","갈기 편하게 미리 손질해둔 레몬은 그날이 지나면 더 이상 갈아서 팔 수가 없습니다..<br>\n","물론, 하루이틀쯤 지났다고 해서 맛이 크게 변할리 없고 냉장고 안에서 쉽게 상할리 없지만, 분명 퀄리티가 다릅니다.<br>\n","오늘도 당신은, 눈물을 흘리며 레몬을 버립니다..."]},{"metadata":{"id":"7rSM3VRjDJn4","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["data = pd.read_csv('https://raw.githubusercontent.com/blackdew/dplus-tensorflow/master/csv/Lemonade.csv')\n","print(data.shape)\n","print(data.describe())\n","data.head(n=10)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"GPJ9upm7DJn_","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# Pandas Dataframe to Numpy Array\n","np_data = data.values\n","\n","# Shuffle\n","np.random.shuffle(np_data)\n","\n","# Split: train - 1600개, test - 나머지\n","train, test = np_data[:1600], np_data[1600:]"],"execution_count":0,"outputs":[]},{"metadata":{"id":"MMHvYEMkJdB8","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["X_train, Y_train = train[:, 0].reshape(-1, 1), train[:, 1].reshape(-1, 1)\n","X_test, Y_test = test[:, 0].reshape(-1, 1), test[:, 1].reshape(-1, 1)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"dmZQKZlMDJoD","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# 20개의 데이터 산점도\n","data[:20].plot.scatter(x='Temperature', y='lemonade')"],"execution_count":0,"outputs":[]},{"metadata":{"id":"ufPSnSIfDJoH","colab_type":"text"},"cell_type":"markdown","source":["## 02. 손으로 그려보는 Linear Regression\n","\n","\n","\n","*직접, 손으로 구조를 설계 해보는 것은 중요하다.*<br>\n","\n","여러분이 텐서플로우로 '그래프'를 코딩하기에 최대한 편한 설계도를 그리면 된다.<br>\n","꼭 Data flow Graph, Computation Grpah, Probabilistic Graph 등의 표현을 따를 필요는 없다.(물론, 익히면 좋다.)<br>\n","#### 이번엔 같이 그려보자!\n","##### 하나하나 이해하면서 다같이 그려보자!\n","\n","![regression](https://github.com/blackdew/dplus-tensorflow/blob/master/img/regression_0.PNG?raw=true)"]},{"metadata":{"id":"bLlTJ8teDJoI","colab_type":"text"},"cell_type":"markdown","source":["#### 차원을 반드시 고려 해야한다."]},{"metadata":{"id":"7np8GnwfDJoJ","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["print(X_train.shape, Y_train.shape)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"LwnU_HKMDJoN","colab_type":"text"},"cell_type":"markdown","source":["## 03. Make a Graph with TF"]},{"metadata":{"id":"GJMY-0dhDJoO","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["'''\n","print(X_train.shape, Y_train.shape)\n","(1600, 1),   (1600, 1)\n","'''\n","# Graph Clear & Make your Graph reproducible\n","tf.reset_default_graph()\n","tf.set_random_seed(2017)\n","\n","# Hyper-Parameters & Option\n","learning_rate = 0.01\n","\n","# Placeholders\n","X = tf.placeholder(dtype=tf.float32, shape=[None, 1])\n","Y = tf.placeholder(dtype=tf.float32, shape=[None, 1])\n","\n","# Weight & Bias\n","W = tf.Variable(tf.random_normal([1, 1]))\n","b = tf.Variable(tf.random_normal([1]))\n","\n","# Hypothesis\n","Y_pred = tf.add(tf.matmul(X, W), b)\n","\n","# Cost(loss) function & Optimizer\n","cost = tf.reduce_mean(tf.squared_difference(Y, Y_pred))\n","optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"eLAZC7LhDJoS","colab_type":"text"},"cell_type":"markdown","source":["## 04. Train! Session!"]},{"metadata":{"id":"d_q7sfsODJoT","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["sess = tf.Session()\n","sess.run(tf.global_variables_initializer())\n","\n","learning_rate = 1\n","\n","start = time.clock()\n","for i in range(2000):\n","    # 학습 진행\n","    sess.run(optimizer, feed_dict={X: X_train, Y: Y_train})\n","    \n","    if (i % 100 == 0):\n","        # 학습 상황 디스플레이\n","        tr_l = sess.run(cost, feed_dict={X: X_train, Y: Y_train})\n","        print(\"{}th training---- training loss: {:.2f}\".format(i, tr_l))\n","\n","print(\"Training Complete\")\n","end = time.clock()\n","\n","print(\"Total elapsed time: {}\".format(end - start))\n","print(\"Test cost(loss): {:.2f}\".format(sess.run(cost, feed_dict={X: X_test, Y: Y_test})))\n","\n","Y_predicted = sess.run(Y_pred, feed_dict={X: X_test, Y: Y_test})"],"execution_count":0,"outputs":[]},{"metadata":{"id":"W65Zp7WoDJoY","colab_type":"text"},"cell_type":"markdown","source":["#### 데이터의 인풋은 온도였습니다. 아웃풋은 판매량이었구요.\n","\n","아래의 내용이 해석이 되어야 합니다."]},{"metadata":{"id":"1T3YKSGUDJoY","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["print(sess.run([W, b]))"],"execution_count":0,"outputs":[]},{"metadata":{"id":"2J8Mt_8QDJob","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# 세션을 닫아야 합니다.\n","sess.close()"],"execution_count":0,"outputs":[]},{"metadata":{"id":"64hkqrjWDJoe","colab_type":"text"},"cell_type":"markdown","source":["### Check"]},{"metadata":{"id":"xYSMO2EwDJof","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["temp = pd.DataFrame({'Y_real': Y_test.reshape(-1, ), 'Y_predicted': Y_predicted.reshape(-1, )})\n","temp.plot.scatter(x='Y_real', y='Y_predicted', figsize=(5, 5))"],"execution_count":0,"outputs":[]},{"metadata":{"id":"x5X3wANMDJok","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["%reset()"],"execution_count":0,"outputs":[]},{"metadata":{"id":"NWex6ndYDJop","colab_type":"text"},"cell_type":"markdown","source":["# Multivariate Linear Regression"]},{"metadata":{"id":"cBvoj4CSDJoq","colab_type":"text"},"cell_type":"markdown","source":["## 00. 라이브러리 불러오기"]},{"metadata":{"id":"wxfbLkxADJor","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["%matplotlib inline\n","%config InlineBackend.figure_format = 'retina'\n","\n","import pandas as pd\n","import numpy as np\n","import tensorflow as tf\n","import matplotlib.pyplot as plt\n","import time"],"execution_count":0,"outputs":[]},{"metadata":{"id":"HK_W_6fzDJou","colab_type":"text"},"cell_type":"markdown","source":["## 01. 데이터 불러오기 & 문제정의 & 전처리"]},{"metadata":{"id":"t11bf8x2DJou","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["rides = pd.read_csv('https://raw.githubusercontent.com/blackdew/dplus-tensorflow/master/csv/Bike-Sharing-Hour.csv')\n","print(rides.shape)\n","rides.head(n=3)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"-4l33evsDJox","colab_type":"text"},"cell_type":"markdown","source":["#### 데이터 설명.\n","\n","Bike sharing systems are new generation of traditional bike rentals where whole process from membership, rental and return back has become automatic.<br>\n","(구글번역) 자전거 공유 시스템은 멤버쉽, 렌탈 및 리턴 백의 전체 프로세스가 자동으로 이루어지는 전통적인 자전거 렌탈의 차세대 제품입니다.<br>\n","\n","[들어가서 Data Set Information을 읽어보자.](http://archive.ics.uci.edu/ml/datasets/Bike+Sharing+Dataset)\n","\n","[ 무언가를 위한 어떤 필요!! ]\n","의 첫 출발로 자전거를 빌려가는 양에 대한 예측을 진행할 것이다."]},{"metadata":{"id":"EMespsDYDJoy","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# 일부의 데이터를 그래프로 확인.\n","rides[:24*10].plot(x='dteday', y='cnt')"],"execution_count":0,"outputs":[]},{"metadata":{"id":"ov13SS1JDJo0","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["rides[:24*10].plot.scatter(x='atemp', y='cnt')"],"execution_count":0,"outputs":[]},{"metadata":{"id":"yZVi6k5eDJo5","colab_type":"text"},"cell_type":"markdown","source":["### Dummy Variable & Drop\n","\n","데이터셋 안에 '국가'라는 컬럼이 있다고 생각해보자.<br>\n","그걸 그대로 사용할 때 펼쳐질 에러에 대해서 논해보자."]},{"metadata":{"id":"86X8FvT9DJo7","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# 더비 변수 생성, one hot encoding\n","dummy_fields = ['season', 'weathersit', 'mnth', 'hr', 'weekday']\n","for each in dummy_fields:\n","    dummies = pd.get_dummies(rides[each], prefix=each, drop_first=False)\n","    rides = pd.concat([rides, dummies], axis=1)\n","\n","# 필요없는 칼럼 제거\n","fields_to_drop = ['instant', 'dteday', 'season', 'weathersit', 'mnth', 'hr', 'weekday']\n","data = rides.drop(fields_to_drop, axis=1)\n","data.head()"],"execution_count":0,"outputs":[]},{"metadata":{"id":"3xE7u-rUDJo-","colab_type":"text"},"cell_type":"markdown","source":["### Scaling\n","결과값 검증할 때에는 scailing했던 내용을 반대로 적용(**unscailing**)해서 검증해야 한다. <br>\n","아래 쪽에 **unscaling**을 위한 코드 부분이 나올 때에 비교해서 살펴보도록 하자."]},{"metadata":{"id":"3PMOvehaDJpA","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# scaling 이전의 데이터\n","data[['casual', 'registered', 'cnt', 'temp', 'atemp', 'hum', 'windspeed']].head(n=10)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"VrNvkOP3DJpD","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["quant_features = ['casual', 'registered', 'cnt', 'temp', 'atemp', 'hum', 'windspeed']\n","# Store scalings in a dictionary so we can convert back later\n","scaled_features = {}\n","for each in quant_features:\n","    mean, std = data[each].mean(), data[each].std()\n","    scaled_features[each] = [mean, std]\n","    data.loc[:, each] = (data[each] - mean) / std\n","    \n","data[['casual', 'registered', 'cnt', 'temp', 'atemp', 'hum', 'windspeed']].head(n=10)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"pp25YlrXDJpH","colab_type":"text"},"cell_type":"markdown","source":["### Splitting the data into training, testing sets\n","\n","맨 마지막의 61일을 test_set으로, 나머지를 Training set으로 둔다.<br>\n","\"일단은 익숙해지자. \""]},{"metadata":{"id":"9iYCMla5DJpI","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["train_data, test_data = data[:-61*24], data[-61*24:]\n","\n","target_fields = ['cnt', 'casual', 'registered']\n","test_features, test_targets = test_data.drop(target_fields, axis=1), test_data[target_fields]\n","train_features, train_targets = train_data.drop(target_fields, axis=1), train_data[target_fields]\n","\n","train_targets.head()"],"execution_count":0,"outputs":[]},{"metadata":{"id":"z60IFXa2DJpQ","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["test_X, test_Y = test_features.values, test_targets.values\n","train_X, train_Y = train_features.values, train_targets.values\n","\n","print(train_X, train_Y)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"heHocNj0DJpT","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["print(train_X.shape, train_Y.shape)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"p691kMhFDJpW","colab_type":"text"},"cell_type":"markdown","source":["## 02. 손으로 그려보는 Linear Regression\n","\n","\n","\n","*직접, 손으로 구조를 설계 해보는 것은 중요하다.*<br>\n","\n","여러분이 텐서플로우로 '그래프'를 코딩하기에 최대한 편한 설계도를 그리면 된다.<br>\n","꼭 Data flow Graph, Computation Grpah, Probabilistic Graph 등의 표현을 따를 필요는 없다.(물론, 익히면 좋다.)<br>"]},{"metadata":{"id":"KOu1Al_VDJpX","colab_type":"text"},"cell_type":"markdown","source":["### 예시.  여러분이 직접 그려보기 전에는 실행 시키지 말기!\n","\n","[리그레션01](https://github.com/blackdew/dplus-tensorflow/blob/master/img/regression_1.PNG?raw=true)\n","[리그레션02](https://github.com/blackdew/dplus-tensorflow/blob/master/img/regression_2.PNG?raw=true)"]},{"metadata":{"id":"iugaCRJsDJpY","colab_type":"text"},"cell_type":"markdown","source":["### 반드시 차원을 고려 할 것!\n","\n","X의 shape = [ 15915 , 58 ]\n","\n","Y의 shape = [ 15915 , 3 ]\n","\n","W의 shape = [ 58 , 3 ]\n","\n","bias의 shape = [ 3 ]\n","\n","연산순서는?"]},{"metadata":{"id":"BtpBU9dVDJpZ","colab_type":"text"},"cell_type":"markdown","source":["## 03. Make a Graph with TF"]},{"metadata":{"id":"bgpIe9C0DJpa","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["'''\n","여러분이 직접 스크립트를 짜야 합니다.\n","조금 힌트는 드립니다..!\n","\n","print(train_X.shape, train_Y.shape)\n","(15435, 58),   (15435, 1)\n","'''\n","# Graph Clear & Make your Graph reproducible\n","tf.reset_default_graph()\n","tf.set_random_seed(2017)\n","\n","# Hyper-Parameters & Option\n","learning_rate = 0.01\n","\n","# Placeholders\n","X = tf.placeholder(dtype=tf.float32, shape=[None, 58])\n","Y = tf.placeholder(dtype=tf.float32, shape=[None, 3])\n","\n","# Weight & Bias\n","W = tf.Variable(tf.random_normal([58, 3]))\n","b = tf.Variable(tf.random_normal([3]))\n","\n","# Hypothesis\n","Y_pred = tf.add(tf.matmul(X, W), b)\n","\n","# Cost(loss) function & Optimizer\n","cost = tf.reduce_mean(tf.squared_difference(Y, Y_pred))\n","Optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"IN9CIiZ5DJpg","colab_type":"text"},"cell_type":"markdown","source":["## 04. Train! Session!"]},{"metadata":{"id":"dWZpn8vjDJph","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["'''\n","Weight, bias 초기화를 먼저 해야함.\n","\n","초기화란? :\n","'''\n","sess = tf.Session()\n","sess.run(tf.global_variables_initializer())\n","\n","learning_rate = 0.01\n","\n","start = time.clock()\n","for i in range(2000):\n","    # 학습 진행\n","    sess.run(Optimizer, feed_dict={X: train_X, Y: train_Y})\n","    \n","    if (i % 100 == 0):\n","        # 학습 상황 디스플레이\n","        tr_l = sess.run(cost, feed_dict={X: train_X, Y: train_Y})\n","        print(\"{}th training---- training loss: {:.2f}\".format(i, tr_l))\n","        \n","print(\"Training Complete\")\n","end = time.clock()\n","\n","print(\"Total elapsed time: {}\".format(end - start))\n","print(\"Test loss: {:.2f}\".format(sess.run(cost, feed_dict={X: test_X, Y: test_Y})))\n","\n","Y_predicted = sess.run(Y_pred, feed_dict={X: test_X, Y: test_Y})"],"execution_count":0,"outputs":[]},{"metadata":{"id":"Octa3y2uDJpo","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["weight, bias = sess.run([W, b])\n","print(weight)\n","print(bias)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"qDhWUQ3qDJps","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# 세션을 닫아야 합니다.\n","sess.close()"],"execution_count":0,"outputs":[]},{"metadata":{"id":"-F19r0R5LMYH","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["print(scaled_features)\n","column = 'cnt'\n","cindex = target_fields.index(column)\n","print(cindex)\n","\n","print(Y_predicted[:, 0] * scaled_features['cnt'][1] + scaled_features['cnt'][0])"],"execution_count":0,"outputs":[]},{"metadata":{"id":"vw_IDP2jDJpv","colab_type":"text"},"cell_type":"markdown","source":["### Check Out on Test set\n","**unscailing** 코드 확인"]},{"metadata":{"id":"SMkY0r6VDJpv","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# target_fields = ['cnt', 'casual', 'registered']\n","column = 'cnt'\n","cindex = target_fields.index(column)\n","rows = 100\n","\n","# 예측값도 Scaling 되어 있으므로 그것을 원래대로 되돌려주는 과정\n","mean, std = scaled_features[column]\n","simple_predictions = Y_predicted[:rows, cindex] * std + mean\n","simple_predictions\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"qtidjRcuLTf6","colab_type":"text"},"cell_type":"markdown","source":["#### 그래프를 그려보자 "]},{"metadata":{"id":"NwfJUSFGDJp0","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["fig, ax = plt.subplots(figsize=(10, 5))\n","\n","ax.plot(simple_predictions, label='Simple_Prediction')\n","ax.plot(test_Y[:rows, cindex] * std + mean, label='Real')\n","ax.set_xlim(right=len(simple_predictions))\n","ax.legend()\n","\n","dates = pd.to_datetime(rides.loc[test_data[:rows].index]['dteday'])\n","dates = dates.apply(lambda d: d.strftime('%b %d'))\n","ax.set_xticks(np.arange(len(dates))[12::24])\n","_ = ax.set_xticklabels(dates[12::24], rotation=45)"],"execution_count":0,"outputs":[]},{"metadata":{"id":"Vq7JpUi8Lbpn","colab_type":"text"},"cell_type":"markdown","source":["### 실습"]},{"metadata":{"id":"UQiG6FHRLexB","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["import pandas as pd \n","\n","# 1. 'https://raw.githubusercontent.com/blackdew/dplus-tensorflow/master/csv/Bike-Sharing-Hour.csv' 데이터를 읽어들인다.\n","rides = pd.read_csv('https://raw.githubusercontent.com/blackdew/dplus-tensorflow/master/csv/Bike-Sharing-Hour.csv')\n","\n","# 2. ['season', 'weathersit', 'mnth', 'hr', 'weekday'] 칼럼은 더미변수화 한다.\n","dummy_fields = ['season', 'weathersit', 'mnth', 'hr', 'weekday']\n","for each in dummy_fields:\n","    dummies = pd.get_dummies(rides[each], prefix=each, drop_first=False)\n","    rides = pd.concat([rides, dummies], axis=1)\n","    \n","# 3. 필요 없어진 ['instant', 'dteday', 'season', 'weathersit', 'mnth', 'hr', 'weekday'] 칼럼을 제거한다.\n","fields_to_drop = ['instant', 'dteday', 'season', 'weathersit', 'mnth', 'hr', 'weekday']\n","data = rides.drop(fields_to_drop, axis=1)\n","data.head()\n","\n","# 4. ['casual', 'registered', 'cnt', 'temp', 'atemp', 'hum', 'windspeed'] 칼럼을 scailing 한다. \n","quant_features = ['casual', 'registered', 'cnt', 'temp', 'atemp', 'hum', 'windspeed']\n","# Store scalings in a dictionary so we can convert back later\n","scaled_features = {}\n","for each in quant_features:\n","    mean, std = data[each].mean(), data[each].std()\n","    scaled_features[each] = [mean, std]\n","    data.loc[:, each] = (data[each] - mean) / std\n","    \n","# 5. test set과 training set로 분리한다.\n","train_data, test_data = data[:-61*24], data[-61*24:]\n","\n","target_fields = ['cnt', 'casual', 'registered']\n","test_features, test_targets = test_data.drop(target_fields, axis=1), test_data[target_fields]\n","train_features, train_targets = train_data.drop(target_fields, axis=1), train_data[target_fields]\n","\n","# 6. ['cnt', 'casual', 'registered'] 칼럼의 값을 target value로 지정한다.\n","test_X, test_Y = test_features.values, test_targets.values\n","train_X, train_Y = train_features.values, train_targets.values\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"Qy6uKabHLgO0","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["'''\n","TensorFlow 그래프를 전체를 직접 작성해 보자\n","'''\n","import tensorflow as tf \n","\n","# Graph Clear & Make your Graph reproducible\n","tf.reset_default_graph()\n","tf.set_random_seed(2017)\n","\n","# Hyper-Parameters & Option\n","learning_rate = 0.01\n","\n","# Placeholders\n","X = tf.placeholder(dtype=tf.float32, shape=[None, 58])\n","Y = tf.placeholder(dtype=tf.float32, shape=[None, 3])\n","\n","# Weight & Bias\n","W = tf.Variable(tf.random_normal([58, 3]))\n","b = tf.Variable(tf.random_normal([3]))\n","\n","# Hypothesis\n","Y_pred = tf.add(tf.matmul(X, W), b)\n","\n","# Cost(loss) function & Optimizer\n","cost = tf.reduce_mean(tf.squared_difference(Y, Y_pred))\n","Optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"-_cicDO8Lhds","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["'''\n","학습하는 세션 코드 전체를 직접 작성해 보자\n","'''\n","sess = tf.Session()\n","sess.run(tf.global_variables_initializer())\n","\n","learning_rate = 0.01\n","\n","for i in range(1000):\n","    # 학습 진행\n","    sess.run(Optimizer, feed_dict={X: train_X, Y: train_Y})\n","    \n","    if (i % 100 == 0):\n","        # 학습 상황 디스플레이\n","        tr_l = sess.run(cost, feed_dict={X: train_X, Y: train_Y})\n","        print(\"{}th training---- training loss: {:.2f}\".format(i, tr_l))\n","        \n","print(\"Training Complete\")\n","\n","print(\"Test loss: {:.2f}\".format(sess.run(cost, feed_dict={X: test_X, Y: test_Y})))\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"LdibgkMyDJp3","colab_type":"text"},"cell_type":"markdown","source":["# must Summary!"]}]}