# -*- coding: utf-8 -*- """ Created on Fri Jun 03 21:00:28 2016 @author: hardy_000 """ import matplotlib.pyplot as plt import pandas as pd import numpy as np import matplotlib as mpl import xgboost as xgb from mpl_toolkits.mplot3d import Axes3D from sklearn import neighbors, datasets from sklearn.neighbors import NearestNeighbors from sklearn import cross_validation from sklearn.metrics import pairwise from scipy.spatial import distance import scipy as sp from sklearn import decomposition import sklearn as skl from collections import Counter from sklearn import preprocessing fw = [500, 1000, 4, 3, 2, 10] def loadtraindata(): z= pd.read_csv("../input/train.csv") return z def loadtestdata(): z= pd.read_csv("../imput/test.csv") return z def removeNonRecentPlaces(df): s=df.place_id[~np.in1d(np.unique(np.where(df.year==1)),np.unique(np.where(df.year==2)))] df=df[np.in1d(df.place_id,s)] return df def convertTime(df): df['x']=df['x'] df['y']=df['y'] initial_date = np.datetime64('2014-01-01T01:01', dtype='datetime64[m]') d_times = pd.DatetimeIndex(initial_date + np.timedelta64(int(mn), 'm') for mn in df.time.values) df['hour'] = (d_times.hour+ d_times.minute/60) df['weekday'] = d_times.weekday df['month'] = d_times.month df['year'] = (d_times.year - 2013) df = df.drop(['time'], axis=1) return df def scaleFeatures(df): df['x']=df['x']*fw[0] df['y']=df['y']*fw[1] df['hour']=df['hour']*fw[2] df['weekday']=df['weekday']*fw[3] df['month']=df['month']*fw[4] df['year']=df['year']*fw[5] return df def getSquare(data,i,j,buff): data=data[((((j-1-buff)<=data['y']) & (data['y']<=j+buff))&(((i-1-buff)<=data['x']) & (data['x']<=i+buff)))] return data def removeSquare(data,i,j): data=data[~((((j-1)<=data['y']) & (data['y']<=j))&(((i-1)<=data['x']) & (data['x']<=i)))] return data def getPercentError(guess,label): correct=(guess==label) return (1./correct.shape[0])*sum(correct) #places=train.groupby('place_id') #meanplaces=places.mean() #meanplaces=meanplaces.reset_index(drop=False) #plot a few locations means #meanplaces[:100].plot(kind='scatter',x='x',y='y',c='place_id',colormap='RdYlGn') def getSamplePlaces(data,numberOfPlaces=1): return data[data.place_id.isin((data.place_id).unique()[1:numberOfPlaces+1])] def getSamplePlace(numberOfPlaces=1): return train[train.place_id==(train.place_id).unique()[numberOfPlaces]] def normalize(series): return (series-min(series))/(max(series)-min(series)) def myf(a,w): lookupTable, indexed_dataSet = np.unique(a, return_inverse=True) y= np.bincount(indexed_dataSet,w) lookupTable[y.argsort()] res=(lookupTable[y.argsort()][::-1][:3]) ret=np.empty((3)) ret.fill(res[-1]) ret[0:res.shape[0]]=res return ret def trainData(X,y): X_train=X y_train=y le=preprocessing.LabelEncoder() le.fit(y_train) labels=le.transform(y_train) dm_train = xgb.DMatrix(X_train, label=labels) param = {'max_depth':3, 'eta':1, 'silent':1, 'objective':'multi:softprob','num_class': len(le.classes_) } #clf = neighbors.KNeighborsClassifier(25) #clf = neighbors.KNeighborsClassifier(n_neighbors=25, weights='distance', # metric='manhattan') clf=xgb.train(param,dm_train,10) return clf def predictTest(test,clf,y_train,X_train,X_train_acc): distances,indices=clf.kneighbors(test) knearest_locs=np.array(X_train)[indices] knearest_labels=np.array(y_train)[indices] knearest_acc=np.array(X_train_acc)[indices] points=np.array(test) reppoints=np.tile(points,(25,1,1)) reppoints=reppoints.transpose((1,0,2)) ty=knearest_locs-reppoints tys=np.square(ty) tyssum=np.sum(tys,axis=2) tyssums=np.sqrt(tyssum) tyssumsac=knearest_acc/knearest_acc result = np.empty_like(knearest_labels[:,0:3]) for i,(x,y) in enumerate(zip(knearest_labels,tyssumsac)): result[i] = myf(x,y) return result def predictRegion(train,test): #train['x']=normalize(train['x']) #train['y']=normalize(train['y']) #train['time']=normalize(train['time']) train['accuracy']=normalize(train['accuracy']) train=scaleFeatures(train) test=scaleFeatures(test) #test['x']=normalize(test['x']) #test['y']=normalize(test['y']) #test['time']=normalize(test['time']) y_train=train.place_id X_train=train.drop('place_id',1) X_train_acc=X_train.accuracy X_train=X_train.drop('accuracy',1) X_train=X_train.drop('row_id',1) ft=trainData(X_train,y_train) X_test=test.drop('accuracy',1) row_id=X_test['row_id'] X_test=X_test.drop('row_id',1) #pred=predictTest(X_test,ft,y_train,X_train,X_train_acc) pred=ft.predict(xgb.DMatrix(X_test))[:,0:3] result=np.insert(pred,0,row_id,axis=1) return result #kmeans starts here normalize data #train=getSquare(train,1,1) #train=getSquare(loadtraindata(),1,1,0.3) train, test = cross_validation.train_test_split(getSquare(loadtraindata(),1,1,0.3), test_size=0.33, random_state=42) testing=True train=convertTime(train) test=convertTime(test) train=removeNonRecentPlaces(train) gridsize=11 if(testing==True): gridsize=1 #train=train.drop('time',1) #nd=normalize(small['accuracy']) #small=small[nd<0.1] pred= np.empty((0,4), int) labels=np.empty((0,),int) for i in range(1,gridsize+1): for j in range(1,gridsize+1): testSquare=getSquare(test,i*0.1,j*0.1,0) if(testing==True): labels=np.append(labels,np.array(testSquare.place_id),0) testSquare=testSquare.drop('place_id',axis=1) res=predictRegion(getSquare(train,i*0.1,j/5*0.1,0.3*0.1),testSquare) test=removeSquare(test,i,j) pred=np.append(pred,res,0) indexOrder=np.argsort(pred[:,0]) pred=pred[indexOrder] if(testing==True): print(getPercentError(pred[:,1],labels[indexOrder])) re=pd.DataFrame(pred) x=re[1].astype(str)+" "+re[2].astype(str)+" "+re[3].astype(str) re=pd.DataFrame(np.concatenate((pred[:,0].reshape(-1,1),x.reshape(-1,1)),1)) re.columns=["row_id","place_id"] re.to_csv("submission.csv",index=False) data = np.random.rand(5,10) # 5 entities, each contains 10 features label = np.random.randint(2, size=5) # binary target dtrain = xgb.DMatrix( data, label=label)