import numpy as np import pandas as pd import re import csv from random import random from time import time from sklearn.neighbors import KNeighborsClassifier titlelist = {"Mr":0, \ "Ms":1, "Miss":1, "Mlle":1,\ "Mrs":2, "Master":2, "Mme":2,\ "Col":3, "Major":3, "Capt":3, "Rev":3, "Dr":3,\ "Sir":4, "Don":4, "Jonkheer":4, "Lady":4,"the Countess":4} #Print you can execute arbitrary python code train = pd.read_csv("../input/train.csv", dtype={"Age": np.float64}, ) test = pd.read_csv("../input/test.csv", dtype={"Age": np.float64}, ) #Print to standard output, and see the results in the "log" section below after running your script print("\n\nTop of the training data:") print(train.head()) print("\n\nSummary statistics of training data") print(train.describe()) #Any files you save will be available in the output tab below #train.to_csv('copy_of_the_training_data.csv', index=False) def processRow(row,titlelist=titlelist): sexList = {"male":0, "female":1} embarkList = {"C":1, "Q":2, "S":3} features = list() label = np.array(int(row[1])) #label PassengerId = np.array(int(row[0])) #id title = re.findall(', (.*?)\. ',row[3])[0] features.append(int(row[2])) #Pclass try: features.append(titlelist[title]) #Title except: features.append(0) #Title features.append(sexList[row[4]]) #Sex try: features.append(int(row[5])) #Age except: features.append(28) #Median of Age features.append(int(row[6])) #Siblings features.append(int(row[7])) #Parents features.append(float(row[9])) #Fare features.append(embarkList.get(row[11],0)) #Embark Place features.append(\ float(row[9])/(1.0+float(row[6])+float(row[7]))) #Fare per person try: features.append(int(row[2])*int(row[5])) #Class*Age except: features.append(int(row[2])*28) features.append(sexList[row[4]]*int(row[2])) #Sex*class return features+[label,PassengerId] def processData(datafile="train.csv"): features = list() with open(datafile,"r") as csvfile: reader = csv.reader(csvfile, delimiter=",") next(reader, None) for row in reader: features.append(processRow(row)) features = np.array(features) labels = features[:,-2] PassengerIds = features[:,-1] features = features[:,:-2] avg = np.average(features,axis=0) maxmin = np.max(features,axis=0)-np.min(features,axis=0) features = (features-avg)/maxmin #print maxmin return features, labels,PassengerIds,avg, maxmin def predictData(avg,maxmin,datafile="test.csv"): features = list() with open(datafile,"r") as csvfile: reader = csv.reader(csvfile, delimiter=",") next(reader, None) for row in reader: features.append(predictRow(row)) features = np.array(features) PassengerIds = features[:,-1] features = features[:,:-1] features = (features-avg)/maxmin #print maxmin return features, PassengerIds def predictRow(row,titlelist=titlelist): sexList = {"male":0, "female":1} embarkList = {"C":1, "Q":2, "S":3} features = list() passendgerID = int(row[0]) #id title = re.findall(', (.*?)\. ',row[2])[0] features.append(int(row[1])) #Pclass try: features.append(titlelist[title]) #Title except: features.append(0) #Title features.append(sexList[row[3]]) #Sex try: features.append(int(row[4])) #Age except: features.append(28) #Median of Age features.append(int(row[5])) #Siblings features.append(int(row[6])) #Parents try: features.append(float(row[8])) #Fare except: features.append(0) features.append(embarkList.get(row[10],0)) #Embark Place try: features.append(\ float(row[8])/(1.0+float(row[5])+float(row[6]))) #Fare per person except: features.append(0) try: features.append(int(row[1])*int(row[4])) #Class*Age except: features.append(int(row[1])*28) features.append(sexList[row[3]]*int(row[1])) #Sex*class return features+[passendgerID] def savePredictions(Ids,predictions,title): try: fh = open("Prediction_"+title+".csv","w") fh.writelines("PassengerId,Survived\n") for idx,row in enumerate(predictions): fh.writelines(str(int(Ids[idx]))+","+str(int(row))+'\n') except: print("Predictions not saved") finally: fh.close() def generateSets(features,labels,cv,test,rnd=True): f_train = list() l_train = list() f_test = list() l_test = list() f_cv = list() l_cv = list() for label,item in enumerate(features): if rnd: value = random() else: value = float(label)/float(len(features)) if value < cv: f_cv.append(item) l_cv.append(labels[label]) elif value < cv+test: f_test.append(item) l_test.append(labels[label]) else: f_train.append(item) l_train.append(labels[label]) return np.array(f_train), np.array(l_train), \ np.array(f_cv), np.array(l_cv), \ np.array(f_test), np.array(l_test) def createLog(clf,clf_type,time,cv_score,test_score): result = clf.get_params().copy() log = "Top "+clf_type+" classifier is:\n" for item in result.keys(): log += item +": "+str(result[item])+"\n" log += "\nTraining time: "+str(time)+"s\n"+\ "CV score: "+str(cv_score)+"\n"+\ "Test score: "+str(test_score) return log #features, labels = processData("train.csv") '''features_train, labels_train,\ features_cv, labels_cv,\ features_test, labels_test \ = generateSets(features,labels,0.15,0.15) ''' def evaluate(clf, top_score, top_clf,f_t,l_t): score = clf.score(f_t, l_t) if score > top_score: return clf, score else: return top_clf, top_score features, labels, PassengerId, avg, maxmin = processData("../input/train.csv") features_train, labels_train,\ features_cv, labels_cv,\ features_test, labels_test \ = generateSets(features,labels,0.1,0.01,rnd=False) #w = [0.001, 0.003, 0.01, 0.03, 0.1, 0.3, 1.0, 3.0, 10.0, 30.0, 100.0, 200.0, 300.0] w = [0.002, 0.004, 0.006, 0.008, 0.01, 0.012, 0.014, 0.016, 0.018, 0.02] print("kNN") top_general_score = 0.0 top_general_clf = None top_score = 0.0 top_clf = None t0 = time() for n_neighbors in range(2,50):############ for weights in ['uniform', 'distance']: for algorithm in ['auto', 'ball_tree', 'kd_tree', 'brute']: for leaf_size in range(2,50): #### clf_KNN = KNeighborsClassifier(n_neighbors=n_neighbors, weights=weights, \ algorithm=algorithm, \ leaf_size=leaf_size) clf_KNN.fit(features_train, labels_train) top_clf, top_score = \ evaluate(clf_KNN, top_score, top_clf,features_cv,labels_cv) dt = time()-t0 score = top_clf.score(features_test, labels_test) log = createLog(top_clf,"kNN",dt,top_score,score) #sendLog(log) print(log) features_predict, PassengerId_predict = predictData(avg,maxmin,"../input/test.csv") predictions = top_clf.predict(features_predict) #savePredictions(PassengerId_predict,predictions,"kNN")