netflix_prize/competition_code.R at master · warik21/netflix_prize · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
########### Training data (rankings only, no dates):
con = url("http://www.tau.ac.il/~saharon/StatsLearn2020/train_ratings_all.dat")
X.tr = read.table (con)
con = url("http://www.tau.ac.il/~saharon/StatsLearn2020/train_y_rating.dat")
y.tr = read.table (con)

con = url("http://www.tau.ac.il/~saharon/StatsLearn2020/train_y_date.dat")
y.da.tr = read.table (con)

########### Test Data
con = url("http://www.tau.ac.il/~saharon/StatsLearn2020/test_ratings_all.dat")
X.te = read.table (con)

con = url("http://www.tau.ac.il/~saharon/StatsLearn2020/test_y_date.dat")
y.da.te = read.table (con)


con = url("http://www.tau.ac.il/~saharon/StatsLearn2020/movie_titles.txt")
titles = read.table(con,sep=",")
names (X.tr) = substr(as.character(titles[,2]),1,50)


########### Get to know our data a little:
table (y.tr) # What rankings does our target get?
apply(data.frame(X.tr[,1:14],y.tr),2,mean) # Which movies are liked?
cor(y.tr,X.tr[,1:14]) # which movies correlated with Miss Congeniality?
apply (X.tr==0, 2, sum) # how many missing?
cor (y.tr, y.da.tr) # changes with time?


########### Divide training data into training and validation
n = dim(X.tr)[1]
nva = 2000
ntr = n-nva
va.id = sample (n,nva) # choose 2000 points for validation
trtr.date = data.frame (X = X.tr[-va.id,], yda=y.da.tr[-va.id,], y=y.tr[-va.id,]) # include dates
trtr = data.frame (X = X.tr[-va.id,],y=y.tr[-va.id,])
va.date = data.frame (X = X.tr[va.id,], yda=y.da.tr[va.id,], y=y.tr[va.id,]) #include dates
va = data.frame (X = X.tr[va.id,],y=y.tr[va.id,])


########### baseline RMSE
print("the baseline RMSE is:")
sqrt(mean((va$y-mean(trtr$y))^2))


########### RMSE on a full X, filling the 0's with the mean of the column.

###this part is done to calculate the X.tr.full matrix, it takes a while
X.tr.full.col = X.tr
X.te.full.col = X.te
colnames(X.te.full.col) <- colnames(X.tr)

# replace zeros with NA
X.tr.full.col[X.tr.full.col == 0] <- NA
for (i in colnames(X.tr.full.col)){
  #replacing the NA values with the mean for each column
  X.tr.full.col[,i][is.na(X.tr.full.col[,i])] <- mean(X.tr.full.col[,i], na.rm = T)
}

trtr.full.col = data.frame (X = X.tr.full.col[-va.id,],y=y.tr[-va.id,])
va.full.col = data.frame (X = X.tr.full.col[va.id,],y=y.tr[va.id,])


print("the column model results are:")
lin.mod.full.col = lm (y~.,data = trtr.full.col)

print("insample:")
lin.pred.cols.insample = predict(lin.mod.full.col, newdata = trtr.full.col)
lin.pred.cols.insample = pmin(lin.pred.cols.insample, 5)
lin.pred.cols.insample = pmax(lin.pred.cols.insample, 1)
print(sqrt(mean((trtr.full.col$y-lin.pred.cols.insample)^2)))


print("on validation")
lin.pred.cols.va = predict(lin.mod.full.col, newdata = va.full.col)
lin.pred.cols.va = pmin(lin.pred.cols.va, 5)
lin.pred.cols.va = pmax(lin.pred.cols.va, 1)
print(sqrt(mean((va.full.col$y-lin.pred.cols.va)^2)))


########### RMSE on a full X, filling the 0's with the mean of the row.

X.tr.full.row = X.tr
X.te.full.row = X.te
colnames(X.te.full.row) <- colnames(X.tr.full.row)
row.means <- rep(NaN, nrow(X.tr.full.row))

for (j in rownames(X.tr.full.row)){

  #getting the mean without the zeroes:
  curr.row <- data.frame(X.tr.full.row[j ,])
  curr.row <- t(curr.row)
  curr.row <- curr.row[colSums(curr.row != 0) > 0, ]
  row.means[j] <- mean(curr.row)

  for (i in colnames(X.tr.full.row)){

    if (X.tr.full.row[j, i] == 0){
      X.tr.full.row[j, i] <- row.means[j]
    }
    #we use strtoi because we need the integer part of j, which for some reason
    # is not trivial in this stupid language, which thinks "10000"<2931 is True
    #if (strtoi(j) < 2931){
      #if (X.te.full[j, i] == 0){
        #X.te.full[j, i] <- a

      #}
    #}

  }
}

trtr.full.row = data.frame (X = X.tr.full.row[-va.id,],y=y.tr[-va.id,])
va.full.row = data.frame (X = X.tr.full.row[va.id,],y=y.tr[va.id,])


print("the row model results are:")
lin.mod.full.row = lm (y~.,data = trtr.full.row)

print("insample:")
lin.pred.rows.insample = predict(lin.mod.full.row, newdata = trtr.full.row)
lin.pred.rows.insample = pmin(lin.pred.rows.insample, 5)
lin.pred.rows.insample = pmax(lin.pred.rows.insample, 1)
print(sqrt(mean((trtr.full.row$y-lin.pred.rows.insample)^2)))


print("on validation")
lin.pred.rows.va = predict(lin.mod.full.row, newdata = va.full.row)
lin.pred.rows.va = pmin(lin.pred.rows.va, 5)
lin.pred.rows.va = pmax(lin.pred.rows.va, 1)
print(sqrt(mean((va.full.row$y-lin.pred.rows.va)^2)))


########### RMSE on a full X, filling the 0's with the mean of the row if the
########### mean is larger than 4.5, and with the mean of the column otherwise.


X.tr.full.both = X.tr
X.te.full.both = X.te
colnames(X.te.full.both) <- colnames(X.tr.full.both)

for (j in rownames(X.tr.full.both)){

  for (i in colnames(X.tr.full.both)){
    if (X.tr.full.both[j, i] == 0) {
      #he a lover
      if (row.means[j] > 4.5){
        X.tr.full.both[j, i] <- 5
      #he a hater
      } else if (row.means[j] < 1.5){
        X.tr.full.both[j, i] <- 1
      #he is neither
      }else{
        X.tr.full.both[j, i] <- col.means[i]
      }
    }

    #we use strtoi because we need the integer part of j, which for some reason
    # is not trivial in this stupid language, which thinks "10000"<2931 is True
    #if (strtoi(j) < 2931){
    #if (X.te.full[j, i] == 0){
    #X.te.full[j, i] <- a

    #}
    #}

  }
}

trtr.full.both = data.frame (X = X.tr.full.both[-va.id,],y=y.tr[-va.id,])
va.full.both = data.frame (X = X.tr.full.both[va.id,],y=y.tr[va.id,])


print("the both model results for 4.5 are:")
lin.mod.full.both = lm (y~.,data = trtr.full.both)

print("insample:")
lin.pred.both.insample = predict(lin.mod.full.both, newdata = trtr.full.both)
lin.pred.both.insample = pmin(lin.pred.both.insample, 5)
lin.pred.both.insample = pmax(lin.pred.both.insample, 1)
print(sqrt(mean((trtr.full.both$y-lin.pred.both.insample)^2)))


print("on validation")
lin.pred.both.va = predict(lin.mod.full.both, newdata = va.full.both)
lin.pred.both.va = pmin(lin.pred.both.va, 5)
lin.pred.both.va = pmax(lin.pred.both.va, 1)
print(sqrt(mean((va.full.both$y-lin.pred.both.va)^2)))


#trtr = data.frame (X = X.tr[-va.id,], yda=y.da.tr[-va.id,], y=y.tr[-va.id,]) # include dates
trtr.full = data.frame (X = X.tr.full[-va.id,],y=y.tr[-va.id,])
#va = data.frame (X = X.tr[va.id,], yda=y.da.tr[va.id,], y=y.tr[va.id,]) #include dates
va.full = data.frame (X = X.tr.full[va.id,],y=y.tr[va.id,])

lin.mod.full = lm (y~.,data = trtr.full)
lin.pred.full = predict(lin.mod.full, newdata = va.full)
lin.pred.full = pmin(lin.pred.full, 5)
lin.pred.full = pmax(lin.pred.full, 1)
#in sample:
lin.pred.insample = predict(lin.mod.full, newdata = trtr.full)
sqrt(mean((trtr.full$y-lin.pred.insample)^2))
#validation
sqrt(mean((va.full$y-lin.pred.full)^2))


#check everything on test
#not sure whether or not we need a new variable, but I added it for readability
tete.full = data.frame(X = X.te.full)

lin.pred.test = predict(lin.mod.full, newdata = tete.full)
lin.pred.test.cap = pmin(lin.pred.test, 5)
lin.pred.test.cap2 = pmax(lin.pred.test, 1)

final.pred <- lin.pred.test.cap2


#check everything on the virgin data
lin.mod = lm (y~.,data = trtr)
lin.pred = predict(lin.mod, newdata = trtr)
sqrt(mean((trtr$y-lin.pred)^2))