forked from apache/predictionio
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Evaluator.scala
302 lines (267 loc) · 9.76 KB
/
Evaluator.scala
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
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
package io.prediction.engines.stock
import io.prediction.storage.Config
import io.prediction.storage.{ ItemTrend, ItemTrends }
import io.prediction.PIOSettings
import io.prediction.DataPreparator
import io.prediction.Validator
import io.prediction.EvaluatorFactory
import io.prediction.core.AbstractEvaluator
import io.prediction.core.BaseEvaluator
import io.prediction.BaseValidationParams
import io.prediction.BaseValidationResults
import io.prediction.BaseCrossValidationResults
import io.prediction.EmptyParams
import scala.math
// FIXME(yipjustin). Remove ._ as it is polluting the namespace.
import org.saddle._
import org.saddle.index.IndexTime
import scala.collection.mutable.{ Map => MMap }
import com.github.nscala_time.time.Imports._
import breeze.linalg.{ DenseMatrix, DenseVector }
import breeze.stats.{ mean, meanAndVariance }
import nak.regress.LinearRegression
import scala.collection.mutable.ArrayBuffer
import com.twitter.chill.MeatLocker
object StockEvaluator extends EvaluatorFactory {
// Use singleton class here to avoid re-registering hooks in config.
val config = new Config()
val itemTrendsDb = config.getAppdataItemTrends()
override def apply()
: BaseEvaluator[
EvaluationDataParams,
EmptyParams,
TrainingDataParams,
ValidationDataParams,
TrainingData,
Feature,
Target,
Target,
ValidationUnit,
ValidationResults,
BaseCrossValidationResults] = {
//new StockEvaluator
new BaseEvaluator(
classOf[StockDataPreparator],
classOf[StockValidator])
}
}
object LocalFileStockEvaluator extends EvaluatorFactory {
override def apply(): AbstractEvaluator = {
//new StockEvaluator
new BaseEvaluator(
classOf[LocalFileStockDataPreparator],
classOf[StockValidator])
}
}
class LocalFileStockDataPreparator
extends DataPreparator[
EvaluationDataParams,
TrainingDataParams,
ValidationDataParams,
TrainingData,
Feature,
Target] {
def getParamsSet(params: EvaluationDataParams)
: Seq[(TrainingDataParams, ValidationDataParams)] = {
Range(params.fromIdx, params.untilIdx, params.evaluationInterval)
.map(idx => {
val trainParams = new TrainingDataParams(
baseDate = params.baseDate,
untilIdx = idx - 1,
windowSize = params.trainingWindowSize,
marketTicker = params.marketTicker,
tickerList = params.tickerList)
val validationParams = new ValidationDataParams(
baseDate = params.baseDate,
fromIdx = idx,
untilIdx = math.min(
idx + params.evaluationInterval,
params.untilIdx),
marketTicker = params.marketTicker,
tickerList = params.tickerList)
(trainParams, validationParams)
})
}
def prepareTraining(params: TrainingDataParams): TrainingData = {
null
}
def prepareValidation(params: ValidationDataParams)
: Seq[(Feature, Target)] = {
Seq[(Feature, Target)]()
}
}
class StockDataPreparator
extends DataPreparator[
EvaluationDataParams,
TrainingDataParams,
ValidationDataParams,
TrainingData,
Feature,
Target] {
val appid = PIOSettings.appid
val itemTrendsDbGetTicker = StockEvaluator.itemTrendsDb.get(appid, _: String).get
// (predicted, acutal)
def getParamsSet(params: EvaluationDataParams)
: Seq[(TrainingDataParams, ValidationDataParams)] = {
Range(params.fromIdx, params.untilIdx, params.evaluationInterval)
.map(idx => {
val trainParams = new TrainingDataParams(
baseDate = params.baseDate,
untilIdx = idx - 1,
windowSize = params.trainingWindowSize,
marketTicker = params.marketTicker,
tickerList = params.tickerList)
val validationParams = new ValidationDataParams(
baseDate = params.baseDate,
fromIdx = idx,
untilIdx = math.min(
idx + params.evaluationInterval,
params.untilIdx),
marketTicker = params.marketTicker,
tickerList = params.tickerList)
(trainParams, validationParams)
})
}
def getTimeIndex(baseDate: DateTime, marketTicker: String) = {
val spy = itemTrendsDbGetTicker(marketTicker)
val timestamps = spy.daily.map(_._1).distinct.filter(_ >= baseDate)
val index = IndexTime(timestamps: _*)
index
}
def getPriceSeriesFromItemTrend(
timeIndex: IndexTime,
itemTrend: ItemTrend)
: (Series[DateTime, Double], Series[DateTime, Boolean]) = {
// The current implementation imports the whole series and reindex with the
// input timeIndex. Of course not the most efficient one. May revisit later.
val daily = itemTrend.daily
val timestamps: IndexTime = IndexTime(daily.map(_._1): _*)
val aprice = daily.map(_._7).toArray
val active = daily.map(_._8).toArray
val apriceSeries: Series[DateTime, Double] =
Series(Vec(aprice), timestamps).reindex(timeIndex)
val activeSeries = Series(Vec(active), timestamps).reindex(timeIndex)
(apriceSeries, activeSeries)
}
// getData return Option[Series]. It is None when part of the data is not
// clean within timeIndex.
private def getData(timeIndex: IndexTime, ticker: String) = {
val itemTrend = itemTrendsDbGetTicker(ticker)
val (price, active) = getPriceSeriesFromItemTrend(timeIndex, itemTrend)
val allActive = active.values.foldLeft(true)(_ && _)
// Return None if the data is not clean enough for training
(if (allActive) Some(price) else None)
}
def prepareTraining(params: TrainingDataParams): TrainingData = {
val timeIndex = getTimeIndex(params.baseDate, params.marketTicker).slice(
params.untilIdx - params.windowSize, params.untilIdx)
val tickerDataSeq = params.tickerList
.map(ticker => (ticker, getData(timeIndex, ticker)))
.filter { case (ticker, optData) => !optData.isEmpty }
.map { case (ticker, optData) => (ticker, optData.get) }
new TrainingData(price = Frame(tickerDataSeq: _*))
}
// Generate evaluation data set with target data up to idx (exclusive)
// e.g. idx = 5, window_size = 3, data = [3,4,5,6,7,8,9,10,11,12]
// The data visible to this function should be everything up to idx = 5:
// visible_data = [3,4,5,6,7]
// feature use [4,5,6]
// target use [6,7], (as target represents daily return)
def prepareOneValidation(idx: Int,
baseDate: DateTime,
marketTicker: String, tickerList: Seq[String]): (Feature, Target) = {
val featureWindowSize = 60 // engine-specific param
val featureFromIdx = idx - 1 - featureWindowSize
val featureUntilIdx = idx - 1
val timeIndex = getTimeIndex(baseDate, marketTicker)
.slice(featureFromIdx, idx)
// generate (ticker, feature, target)-tuples
val data = tickerList
.map(ticker => (ticker, getData(timeIndex, ticker)))
.filter { case (ticker, optPrice) => !optPrice.isEmpty }
.map {
case (ticker, optPrice) => {
val price = optPrice.get
// feature
val featurePrice = price.slice(0, featureWindowSize)
// target
val todayPrice = price.at(featureWindowSize - 1)
val tomorrowPrice = price.at(featureWindowSize)
val target = math.log(tomorrowPrice) - math.log(todayPrice)
(ticker, featurePrice, target)
}
}
val featureData = Frame(data.map(e => (e._1, e._2)): _*)
val targetData = data.map(e => (e._1, e._3)).toMap
//return (new Feature(data = featureData), new Target(data = targetData))
return (new Feature(
boxedData = MeatLocker(featureData)),
new Target(data = targetData))
}
def prepareValidation(params: ValidationDataParams)
: Seq[(Feature, Target)] = {
(params.fromIdx until params.untilIdx).map(idx =>
prepareOneValidation(idx, params.baseDate,
params.marketTicker, params.tickerList)
).toSeq
}
}
class StockValidator
extends Validator[
EmptyParams,
TrainingDataParams,
ValidationDataParams,
Feature,
Target,
Target,
ValidationUnit,
ValidationResults,
BaseCrossValidationResults] {
def validate(feature: Feature, predicted: Target, actual: Target)
: ValidationUnit = {
val predictedData = predicted.data
val actualData = actual.data
val data = predictedData.map {
case (ticker, pValue) => {
(pValue, actualData(ticker))
}
}.toSeq
new ValidationUnit(data = data)
}
override
def validateSet(
trainingDataParams: TrainingDataParams,
validationDataParams: ValidationDataParams,
validationUnits: Seq[ValidationUnit])
: ValidationResults = {
val results: Seq[(Double, Double)] = validationUnits.map(_.data).flatten
val pThresholds = Seq(-0.01, -0.003, -0.001, -0.0003,
0.0, 0.0003, 0.001, 0.003, 0.01)
val output = pThresholds.map( pThreshold => {
val screened = results.filter(e => e._1 > pThreshold).toSeq
val over = screened.filter(e => (e._1 > e._2)).length
val under = screened.filter(e => (e._1 < e._2)).length
// Sum actual return.
val actuals = screened.map(_._2)
val (mean_, variance, count) = meanAndVariance(actuals)
val stdev = math.sqrt(variance)
// 95% CI
val ci = 1.96 * stdev / math.sqrt(count)
val s = (f"Threshold: ${pThreshold}%+.4f " +
f"Mean: ${mean_}%+.6f Stdev: ${stdev}%.6f CI: ${ci}%.6f " +
f"Total: ${count}%5d Over: $over%5d Under: $under%5d")
println(s)
s
})
new ValidationResults(data = output)
}
override
def crossValidate(
validateResultsSeq
: Seq[(TrainingDataParams, ValidationDataParams, ValidationResults)])
: BaseCrossValidationResults = {
println("Cross Validation")
validateResultsSeq.map(e => e._3.data.map(println))
null
}
}