def multiboxDataTasksSort(self, ct, workers=2, memoryLimit=100, pageSizeOverrideMB=1):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let N = __ct__;
let aPrime = 503
let toSort = Vector.range(N, { ((_ * _) % aPrime, _) }).paged;
let result = cached`(#ExternalIoTask(#DistributedDataOperation(#Sort(toSort))))
sorting.isSorted(result)
""".replace("__ct__", str(ct))
result = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
text,
s3,
workers,
timeout=TIMEOUT,
memoryLimitMb=memoryLimit,
pageSizeOverride=pageSizeOverrideMB*1024*1024
)
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
self.assertTrue(result.asResult.result.pyval == True, result)
def test_sortVecOfVec(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let values = []
let ct = 500000
values = [(ix % 100, Vector.range(40)) for ix in sequence(ct)]
let res = cached`(#ExternalIoTask(#DistributedDataOperation(#Sort(values.paged))));
let firstAreSorted = true;
for ix in sequence(size(res)-1)
if (res[ix][0] > res[ix+1][0])
firstAreSorted = false;
size(res) == size(values) and firstAreSorted
"""
result = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
text, s3, 1, timeout=TIMEOUT, memoryLimitMb=3000)
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
self.assertTrue(result.asResult.result.pyval == True, result)
def test_takeFromLargeObjects(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let N = 100;
//each string is 1 MB
let takeFrom = [" " * 100 * 100 * 10 * 10 + " " * ix for ix in sequence(N)].paged;
let indices = Vector.range(N,fun(x) { x }).paged;
cached`(#ExternalIoTask(#DistributedDataOperation(#Take(indices, takeFrom))))
"""
try:
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
text,
s3,
1,
timeout=TIMEOUT,
memoryLimitMb=1000,
returnSimulation = True,
pageSizeOverride = 1024 * 1024
)
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
for page in result.asResult.result.getVectorPageIds(simulation.getWorkerVdm(0)):
self.assertLess(page.bytecount / 1024.0 / 1024.0, 2.0)
finally:
simulation.teardown()
def test_multiboxDataTasksTake_1(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let N = 10000000;
let isPrime = fun(p) {
let x = 2
while (x*x <= p) {
if (p%x == 0)
return 0
x = x + 1
}
return x
}
let takeFrom = Vector.range(N, isPrime).paged;
let indices = Vector.range(N,fun(x) { (0, (x * 503) % N ) }).paged;
cached`(#ExternalIoTask(#DistributedDataOperation(#Take(indices, takeFrom)))) ==
indices ~~ { takeFrom[_[1]] }
"""
result = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
text,
s3,
2,
timeout=TIMEOUT,
memoryLimitMb=1000
)
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
self.assertTrue(result.asResult.result.pyval == True, result)
def test_splitToRowMajor(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
nRows = 100000
nColumns = 50
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.setupScript(nRows, nColumns),
s3,
1,
timeout=30,
memoryLimitMb=8 * 1024,
threadCount=4,
returnSimulation=True,
useInMemoryCache=False)
try:
self.assertTrue(result.isResult())
setup = result.asResult.result
t0 = time.time()
result = simulation.compute(self.splitToRowMajorScript(),
timeout=360,
chunks=setup)
totalTimeToReturnResult = time.time() - t0
self.assertTrue(result.isResult())
PerformanceTestReporter.recordTest(
"algorithms.text.splitToRowMajor.%srows_%scolumns" %
(nRows, nColumns), totalTimeToReturnResult, None)
finally:
simulation.teardown()
def test_sortHeterogeneous(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let values = []
let ct = 1000000
for ix in sequence(ct)
values = values :: ix :: Float64(ix)
let sortedVals = cached`(#ExternalIoTask(#DistributedDataOperation(#Sort(values.paged))))
let sortedAndHomogenous = fun(v) {
for ix in sequence(size(v)-1)
if (v[ix] >= v[ix+1] or `TypeJOV(v[ix]) is not `TypeJOV(v[ix+1]))
throw (ix, v[ix], v[ix+1])
return true;
}
if (size(sortedVals) != size(values))
throw "expected " + String(size(values)) + ", not " + String(size(sortedVals))
sortedAndHomogenous(sortedVals[,ct]) and
sortedAndHomogenous(sortedVals[ct,])
"""
result = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
text,
s3,
1,
timeout=TIMEOUT,
memoryLimitMb=1000
)
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
self.assertTrue(result.asResult.result.pyval == True, result)
def test_vector_string_apply(self):
#verify that the compiler doesn't crap out during many runs.
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
InMemoryCumulusSimulation.computeUsingSeveralWorkers("""
let v = Vector.range(10000000)
let v2 = v.apply(String)
let v3 = v2.apply({_ + "a"})
v3.sum(size)
""",
s3,
4,
timeout=240)
def test_sortVecOfVec(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let values = []
let ct = 500000
values = [(ix % 100, Vector.range(40)) for ix in sequence(ct)]
let res = cached`(#ExternalIoTask(#DistributedDataOperation(#Sort(values.paged))));
let firstAreSorted = true;
for ix in sequence(size(res)-1)
if (res[ix][0] > res[ix+1][0])
firstAreSorted = false;
size(res) == size(values) and firstAreSorted
"""
result = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
text,
s3,
1,
timeout=TIMEOUT,
memoryLimitMb=3000
)
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
self.assertTrue(result.asResult.result.pyval == True, result)
def basicTaskPathwayTest(self, sz, machines=1, memory=1000):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let N = __size__;
//let values = Vector.range(N,fun(x) { ((x * 503) % N, x) }).paged;
let values = Vector.range(N).paged;
let s1 = cached`(#ExternalIoTask(#DistributedDataOperation(#Sort(values))))
let s2 = sorting.sort(values)
if (size(s1) != size(s2))
return 'wrong size: %s != %s'.format(size(s1), size(s2))
for ix in sequence(size(s1))
if (s1[ix] != s2[ix])
return 'not equal: index=%s. %s != %s'.format(ix, s1[ix], s2[ix])
return true
""".replace("__size__", str(sz))
result = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
text,
s3,
machines,
timeout=TIMEOUT,
memoryLimitMb=memory
)
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
self.assertTrue(result.asResult.result.pyval == True, result)
def test_expansionWithVecOfVec(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
simulation = InMemoryCumulusSimulation.InMemoryCumulusSimulation(
4, #worker count
1,
memoryPerWorkerMB=100,
threadsPerWorker=2,
s3Service=s3)
try:
self.assertTrue(simulation.waitForGlobalScheduler(timeout=2.0))
simulation.getGlobalScheduler().setCheckpointStatusInterval(0.0001)
simulation.submitComputation(
"Vector.range(20, fun(ix) { Vector.range(100000+ix).paged }).paged"
)
simulation.waitForAnyResult()
simulation.addWorker()
self.assertTrue(simulation.waitForHandshake())
finally:
simulation.teardown()
def basic_gpu_works_helper(self, function, onGPU=True):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
testingVectorText = "Vector.range(1024*4, {_+1000000})"
text = """
let f = fun(ct) {
let res = 0.0
let x = 1.0
while (x < ct)
{
x = x + 1.0
res = res + `""" + function + """(x)
}
res
}"""
if onGPU:
text += """`CUDAVectorApply(f,""" + testingVectorText + """)"""
else:
text += testingVectorText + """ ~~ f"""
res = InMemoryCumulusSimulation.computeUsingSeveralWorkers(text, s3, 1, timeout=120, threadCount=4)
self.assertIsNotNone(res)
self.assertTrue(res.isResult(), res)
def test_cumulusCanTriggerNewRegimes(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
simulation = InMemoryCumulusSimulation.InMemoryCumulusSimulation(
4, #worker count
1,
memoryPerWorkerMB=100,
threadsPerWorker=2,
s3Service=s3)
try:
self.assertTrue(simulation.waitForGlobalScheduler(timeout=2.0))
simulation.waitForHandshake()
regime = simulation.getWorker(0).getRegimeHash()
self.assertTrue(regime is not None)
simulation.getWorker(0).triggerRegimeChange()
time.sleep(1.0)
simulation.waitForHandshake()
regime2 = simulation.getWorker(0).getRegimeHash()
self.assertTrue(regime2 is not None)
self.assertTrue(regime2 != regime)
finally:
simulation.teardown()
def basic_gpu_works_helper(self, function, onGPU=True):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
testingVectorText = "Vector.range(1024*4, {_+1000000})"
text = """
let f = fun(ct) {
let res = 0.0
let x = 1.0
while (x < ct)
{
x = x + 1.0
res = res + `""" + function + """(x)
}
res
}"""
if onGPU:
text += """`CUDAVectorApply(f,""" + testingVectorText + """)"""
else:
text += testingVectorText + """ ~~ f"""
res = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
text, s3, 1, timeout=120, threadCount=4)
self.assertIsNotNone(res)
self.assertTrue(res.isResult(), res)
def test_transposeToColumnMajor(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
nRows = 100000
nColumns = 50
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.transposeSetupScript(nRows, nColumns),
s3, 1, timeout = 300, memoryLimitMb = 45 * 1024, threadCount = 30,
returnSimulation = True, useInMemoryCache = False)
try:
self.assertTrue(result.isResult())
rowMajor = result.asResult.result
t0 = time.time()
result = simulation.compute(
self.transposeRowMajorToColumnMajorScript(nRows, nColumns),
timeout = 500,
rowMajor = rowMajor
)
totalTimeToReturnResult = time.time() - t0
self.assertTrue(result.isResult())
PerformanceTestReporter.recordTest(
"algorithms.text.transposeRowMajorToColumnMajor.%srows_%scolumns" % (nRows, nColumns),
totalTimeToReturnResult, None)
finally:
simulation.teardown()
def classSortingTest(self, sz, useClass = True, machines=1, memory=1000):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let N = __size__;
let C = if (__use_class__) { class { member x; } } else { Int64 }
let values = Vector.range(N, C).paged;
let s1 = cached`(#ExternalIoTask(#DistributedDataOperation(#Sort(values))))
return size(s1) == N
""".replace("__size__", str(sz)).replace("__use_class__", '1' if useClass else '0')
result = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
text,
s3,
machines,
timeout=TIMEOUT,
memoryLimitMb=memory
)
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
self.assertTrue(result.asResult.result.pyval == True, result)
def test_vector_string_apply(self):
#verify that the compiler doesn't crap out during many runs.
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
InMemoryCumulusSimulation.computeUsingSeveralWorkers("""
let v = Vector.range(10000000)
let v2 = v.apply(String)
let v3 = v2.apply({_ + "a"})
v3.sum(size)
""",
s3,
4,
timeout=240
)
def gbmRegressionFittingTest(self, nRows, nColumns, depth, nThreads,
maxBoosts):
testName = self.getTestName(nRows, nColumns, depth, maxBoosts,
nThreads)
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.dataGenerationScript(nRows, nColumns),
s3,
1,
timeout=360,
memoryLimitMb=30 * 1024,
threadCount=nThreads,
returnSimulation=True,
useInMemoryCache=False)
try:
self.assertTrue(result.isResult())
dfPredictors, dfResponse = result.asResult.result
fitter = simulation.compute(
self.regressionScript(depth, 1),
timeout=360,
dfResponse=dfResponse,
dfPredictors=dfPredictors).asResult.result
t0 = time.time()
for nBoosts in range(1, maxBoosts):
testName = self.getTestName(nRows, nColumns, depth, nBoosts,
nThreads)
predictions = simulation.compute(
"fitter.predictionsAndPseudoresiduals()",
timeout=360,
fitter=fitter).asResult.result
totalTimeToReturnResult = time.time() - t0
PerformanceTestReporter.recordTest(testName + "_predict",
totalTimeToReturnResult,
None)
fitter = simulation.compute(
"fitter.nextGivenPredictions(predictions)",
timeout=360,
fitter=fitter,
predictions=predictions).asResult.result
totalTimeToReturnResult = time.time() - t0
PerformanceTestReporter.recordTest(testName,
totalTimeToReturnResult,
None)
finally:
simulation.teardown()
def test_CalculationRicochet(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let f = fun(ct, seed = 1) {
let x = 0
let res = []
let it = iterator(math.random.UniformReal(0, size(v), seed))
for ix in sequence(ct) {
let x = Int64(pull it)
res = res :: (x / Float64(size(v)), v[x])
}
return res
}
v[2]
f(__count__,__seed__)
"""
vResult, sim = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
"Vector.range(125000000, math.log)",
s3,
4,
timeout=120,
memoryLimitMb=400,
threadCount=1,
useInMemoryCache=True,
returnSimulation=True)
try:
v = vResult.asResult.result
t0 = time.time()
sim.compute(text.replace("__seed__",
"1").replace("__count__", "1000"),
timeout=120,
v=v)
PerformanceTestReporter.recordTest(
"python.InMemoryCumulus.Ricochet1000.Pass1",
time.time() - t0, None)
t0 = time.time()
sim.compute(text.replace("__seed__",
"2").replace("__count__", "1000"),
timeout=120,
v=v)
PerformanceTestReporter.recordTest(
"python.InMemoryCumulus.Ricochet1000.Pass2",
time.time() - t0, None)
finally:
sim.teardown()
def gbmRegressionFittingTest(self, nRows, nColumns, depth, nThreads, maxBoosts):
testName = self.getTestName(nRows, nColumns, depth, maxBoosts, nThreads)
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.dataGenerationScript(nRows, nColumns),
s3,
1,
timeout = 360,
memoryLimitMb = 30 * 1024,
threadCount = nThreads,
returnSimulation = True,
useInMemoryCache = False
)
try:
self.assertTrue(result.isResult())
dfPredictors, dfResponse = result.asResult.result
fitter = simulation.compute(
self.regressionScript(depth, 1),
timeout = 360,
dfResponse = dfResponse,
dfPredictors = dfPredictors
).asResult.result
t0 = time.time()
for nBoosts in range(1, maxBoosts):
testName = self.getTestName(nRows, nColumns, depth, nBoosts, nThreads)
predictions = simulation.compute(
"fitter.predictionsAndPseudoresiduals()",
timeout = 360,
fitter = fitter
).asResult.result
totalTimeToReturnResult = time.time() - t0
PerformanceTestReporter.recordTest(
testName + "_predict", totalTimeToReturnResult, None)
fitter = simulation.compute(
"fitter.nextGivenPredictions(predictions)",
timeout = 360,
fitter = fitter,
predictions = predictions
).asResult.result
totalTimeToReturnResult = time.time() - t0
PerformanceTestReporter.recordTest(
testName, totalTimeToReturnResult, None)
finally:
simulation.teardown()
def takeTest(indexExpr):
result = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
takeText.replace("__indices__", indexExpr),
s3,
1,
timeout=TIMEOUT,
memoryLimitMb=1000)
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
self.assertTrue(result.asResult.result.pyval == True, result)
def takeTest(indexExpr):
result = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
takeText.replace("__indices__", indexExpr),
s3,
1,
timeout=TIMEOUT,
memoryLimitMb=1000
)
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
self.assertTrue(result.asResult.result.pyval == True, result)
def test_disk_scans(self):
s3 = ActualS3Interface.ActualS3InterfaceFactory()
objectStore = S3ObjectStore.S3ObjectStore(
s3,
Setup.config().userDataS3Bucket,
prefix="test_object_cache/"
)
_, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
"1+1",
s3,
1,
memoryLimitMb=1 * 1024,
threadCount=30,
returnSimulation=True,
ioTaskThreadOverride=8,
objectStore=objectStore,
useInMemoryCache=False #use an actual disk cache for this
)
try:
gigabytes = 8
t0 = time.time()
resultVectors = []
for ix in range(gigabytes):
result = simulation.compute("Vector.range(125000000 + %s)" % ix, timeout=120)
resultVectors.append(result.asResult.result)
t1 = time.time()
intResults = []
for vec in resultVectors:
result = simulation.compute("v.sum()", timeout = 120, v=vec)
intResults.append(result.asResult.result.pyval)
self.assertTrue(len(intResults) == gigabytes)
PerformanceTestReporter.recordTest(
"python.BigBox.Disk.Write.10GB",
t1 - t0,
None
)
PerformanceTestReporter.recordTest(
"python.BigBox.Disk.WriteAndScan.%sGB" % gigabytes,
time.time() - t0,
None
)
finally:
simulation.teardown()
def loadCheckpointFromFreshSimulationTest(self,
calculationText,
timestampsPerPassList,
clientCount=1,
timestep=1.0):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
statuses = []
viewFactory = None
for timestampsThisPass in timestampsPerPassList:
simulation = InMemoryCumulusSimulation.InMemoryCumulusSimulation(
4, #worker count
clientCount,
memoryPerWorkerMB=100,
threadsPerWorker=2,
s3Service=s3,
sharedStateViewFactory=viewFactory)
viewFactory = simulation.sharedStateViewFactory
statusesThisPass = []
try:
self.assertTrue(simulation.waitForGlobalScheduler(timeout=2.0))
simulation.getGlobalScheduler().setCheckpointStatusInterval(
0.1)
for ix in range(clientCount):
simulation.submitComputationOnClient(ix, calculationText)
for subPass in range(timestampsThisPass):
time.sleep(timestep)
statusesThisPass.append(
self.timeElapsedOfMostRecentCheckpoints(simulation))
simulation.getGlobalScheduler(
).triggerFullCheckpointsOnOutstandingComputations()
self.waitForFullCheckpoint(simulation)
statusesThisPass.append(
self.timeElapsedOfMostRecentCheckpoints(simulation))
finally:
for ix in range(4):
simulation.getWorker(ix).dumpStateToLog()
simulation.teardown()
statuses.append(statusesThisPass)
return statuses
def largeDatasetJoinTest(self, mbOfData, columns, threads, machineCount, ratio = .5):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
t0 = time.time()
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.dataGenerationScript(mbOfData, columns),
s3,
machineCount,
timeout = 360,
memoryLimitMb = mbOfData / ratio / machineCount,
#channelThroughputMBPerSecond = 100.0,
threadCount = threads,
returnSimulation = True,
useInMemoryCache = False,
disableEventHandler = True
)
try:
self.assertTrue(result.isResult())
data = result.asResult.result
joinScript = """
let leftDF = dataframe.DataFrame(data[,size(data)/2])
let rightDF = dataframe.DataFrame(data[size(data)/2,])
size(leftDF.join(rightDF, on: "C0", how: `outer, chunkSize: 1000000, areSorted:true))
"""
t0 = time.time()
result = simulation.compute(
joinScript,
timeout=1080,
data=data
)
totalTimeToReturnResult = time.time() - t0
logging.info("Total time to join: %s", totalTimeToReturnResult)
self.assertTrue(result.isResult(), result)
PerformanceTestReporter.recordTest(
"algorithms.Join.inMemory_%sMB_%scols_%sthreads_%smachines" %
(mbOfData, columns,threads,machineCount),
totalTimeToReturnResult,
None
)
finally:
dfResponse = None
dfPredictors = None
result = None
simulation.teardown()
def gbmRegressionFittingTest(self,
nRows,
nColumns,
depth,
nThreads,
nBoosts,
copies,
report=True):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.dataGenerationScript(nRows, nColumns),
s3,
1,
timeout=360,
memoryLimitMb=30 * 1024,
threadCount=nThreads,
returnSimulation=True,
useInMemoryCache=False)
try:
self.assertTrue(result.isResult())
dfPredictors, dfResponse = result.asResult.result
builder = simulation.compute(
self.regressionScript(depth, nBoosts),
timeout=360,
dfResponse=dfResponse,
dfPredictors=dfPredictors).asResult.result
t0 = time.time()
testName = self.getTestName(nRows, nColumns, depth, nBoosts,
nThreads, copies)
result = simulation.compute(
"Vector.range(%s).apply(fun(x) { builder.fit(dfPredictors[,-x-1], dfResponse[,-x-1]) })"
% copies,
timeout=360,
builder=builder,
dfPredictors=dfPredictors,
dfResponse=dfResponse,
).asResult.result
totalTimeToReturnResult = time.time() - t0
if report:
PerformanceTestReporter.recordTest(testName,
totalTimeToReturnResult,
None)
finally:
simulation.teardown()
def test_importanceSampling(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
result = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
importanceSampling,
s3,
4,
memoryLimitMb = 1000,
timeout=240,
useInMemoryCache = False
)
self.assertTrue(result.isResult())
def test_bigLmOnDataframe(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
result = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
importanceSampling,
s3,
1,
memoryLimitMb=4000,
threadCount=4,
timeout=240,
useInMemoryCache=False)
self.assertTrue(result.isResult(), result)
def runOnGPU(self, funcExpr, vecExpr, captureExpr=""):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = captureExpr + """
let f = __funcExpr__;
let vec = __vecExpr__;
cached`(#GpuApply(f, vec));
""".replace("__funcExpr__", funcExpr).replace("__vecExpr__", vecExpr)
res = InMemoryCumulusSimulation.computeUsingSeveralWorkers(text, s3, 1, timeout=120, threadCount=4,memoryLimitMb=1000)
self.assertIsNotNone(res)
self.assertFalse(res.isException(), "Failed with %s" % res)
return res
def largeDatasetJoinTest(self,
mbOfData,
columns,
threads,
machineCount,
ratio=.5):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
t0 = time.time()
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.dataGenerationScript(mbOfData, columns),
s3,
machineCount,
timeout=360,
memoryLimitMb=mbOfData / ratio / machineCount,
#channelThroughputMBPerSecond = 100.0,
threadCount=threads,
returnSimulation=True,
useInMemoryCache=False,
disableEventHandler=True)
try:
self.assertTrue(result.isResult())
data = result.asResult.result
joinScript = """
let leftDF = dataframe.DataFrame(data[,size(data)/2])
let rightDF = dataframe.DataFrame(data[size(data)/2,])
size(leftDF.join(rightDF, on: "C0", how: `outer, chunkSize: 1000000, areSorted:true))
"""
t0 = time.time()
result = simulation.compute(joinScript, timeout=1080, data=data)
totalTimeToReturnResult = time.time() - t0
logging.info("Total time to join: %s", totalTimeToReturnResult)
self.assertTrue(result.isResult(), result)
PerformanceTestReporter.recordTest(
"algorithms.Join.inMemory_%sMB_%scols_%sthreads_%smachines" %
(mbOfData, columns, threads, machineCount),
totalTimeToReturnResult, None)
finally:
dfResponse = None
dfPredictors = None
result = None
simulation.teardown()
def test_bigLmOnDataframe(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
result = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
importanceSampling,
s3,
1,
memoryLimitMb = 4000,
threadCount = 4,
timeout=240,
useInMemoryCache = False
)
self.assertTrue(result.isResult(), result)
def check_precision_of_function_on_GPU(self, function, input):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let f = fun(x) {
`""" + function + """(x)
}
`CUDAVectorApply(f, [""" + str(input) + """])[0]
"""
res = InMemoryCumulusSimulation.computeUsingSeveralWorkers(text, s3, 1, timeout=120, threadCount=4)
self.assertIsNotNone(res)
self.assertTrue(res.isResult(), res)
gpuValue = res.asResult.result.pyval
methodToCall = getattr(math, function)
pythonValue = methodToCall(input)
self.assertTrue(abs(gpuValue - pythonValue) < 1e-10)
def check_precision_of_function_on_GPU(self, function, input):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let f = fun(x) {
`""" + function + """(x)
}
cached`(#GpuApply(f, [""" + str(input) + """]))[0]
"""
res = InMemoryCumulusSimulation.computeUsingSeveralWorkers(text, s3, 1, timeout=120, threadCount=4)
self.assertIsNotNone(res)
self.assertTrue(res.isResult(), res)
gpuValue = res.asResult.result.pyval
methodToCall = getattr(math, function)
pythonValue = methodToCall(input)
self.assertTrue(abs(gpuValue - pythonValue) < 1e-10)
def largeDatasetBigLMTest(self, mbOfData, columns, threads, testName):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
t0 = time.time()
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.dataGenerationScript(mbOfData, columns),
s3,
1,
timeout=360,
memoryLimitMb=50 * 1024,
threadCount=threads,
returnSimulation=True,
useInMemoryCache=False)
if testName is not None:
PerformanceTestReporter.recordTest(testName + "_create",
time.time() - t0, None)
try:
self.assertTrue(result.isResult())
dfResponse, dfPredictors = result.asResult.result
regressionScript = """
let model = math.regression.LinearRegression(dfPredictors, dfResponse, fitIntercept: false);
let coefficients = model.coefficients();
coefficients[0]
"""
t0 = time.time()
result = simulation.compute(regressionScript,
timeout=1080,
dfResponse=dfResponse,
dfPredictors=dfPredictors)
totalTimeToReturnResult = time.time() - t0
self.assertTrue(result.isResult())
if testName is not None:
PerformanceTestReporter.recordTest(testName,
totalTimeToReturnResult,
None)
finally:
dfResponse = None
dfPredictors = None
result = None
simulation.teardown()
def largeDatasetBigLMTest(self, mbOfData, columns, threads, testName):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
t0 = time.time()
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.dataGenerationScript(mbOfData, columns),
s3,
1,
timeout = 360,
memoryLimitMb = 50 * 1024,
threadCount = threads,
returnSimulation = True,
useInMemoryCache = False
)
if testName is not None:
PerformanceTestReporter.recordTest(testName + "_create", time.time() - t0, None)
try:
self.assertTrue(result.isResult())
dfResponse, dfPredictors = result.asResult.result
regressionScript = """
let model = math.regression.LinearRegression(dfPredictors, dfResponse, fitIntercept: false);
let coefficients = model.coefficients();
coefficients[0]
"""
t0 = time.time()
result = simulation.compute(
regressionScript,
timeout=1080,
dfResponse=dfResponse,
dfPredictors=dfPredictors
)
totalTimeToReturnResult = time.time() - t0
self.assertTrue(result.isResult())
if testName is not None:
PerformanceTestReporter.recordTest(testName, totalTimeToReturnResult, None)
finally:
dfResponse = None
dfPredictors = None
result = None
simulation.teardown()
def regressionTreePredictionTest(self,
mbOfData,
columns,
testName,
treeDepth,
threads,
minSamplesSplit=50):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.dataGenerationScript(mbOfData, columns),
s3,
1,
timeout=360,
memoryLimitMb=45 * 1024,
threadCount=threads,
returnSimulation=True,
useInMemoryCache=False)
try:
self.assertTrue(result.isResult())
dfResponse, dfPredictors = result.asResult.result
fitTree = simulation.compute(self.regressionScript(
treeDepth, minSamplesSplit - 1),
timeout=120,
dfResponse=dfResponse,
dfPredictors=dfPredictors)
def predictionScript(dirtyFlag=1):
return ";(%s; fitRegressionTree.predict(dfPredictors));" % dirtyFlag
t0 = time.time()
result = simulation.compute(
predictionScript(),
timeout=120,
dfPredictors=dfPredictors,
fitRegressionTree=fitTree.asResult.result)
totalTimeToReturnResult = time.time() - t0
self.assertTrue(result.isResult())
PerformanceTestReporter.recordTest(testName,
totalTimeToReturnResult, None)
finally:
simulation.teardown()
def test_vector_transpose(self):
#verify that the compiler doesn't crap out during many runs.
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
_, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers("1+1",
s3,
2,
memoryLimitMb = 500,
timeout=10,
returnSimulation = True,
channelThroughputMBPerSecond = 50.0
)
try:
result = simulation.compute("""
let arrangedContiguously = fun (vecs) {
let res = vecs.sum().paged;
let tr = []
let low = 0
for v in vecs {
tr = tr :: res[low,low+size(v)]
low = low + size(v)
}
tr
};
let transpose = fun(vecOfIndexable) {
let vecs = arrangedContiguously(vecOfIndexable)
let n = size(vecs[0]);
[[vecs[jx][ix] for jx in sequence(size(vecs))] for ix in sequence(n)]
};
let vectors = Vector.range(5000, {Vector.range(300)})
transpose(vectors)
""",
timeout = 30.0
)
self.assertTrue(result.isResult())
finally:
simulation.teardown()
def test_vector_transpose(self):
#verify that the compiler doesn't crap out during many runs.
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
_, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers("1+1",
s3,
2,
memoryLimitMb = 500,
timeout=10,
returnSimulation = True,
channelThroughputMBPerSecond = 50.0
)
try:
result = simulation.compute("""
let arrangedContiguously = fun (vecs) {
let res = vecs.sum().paged;
let tr = []
let low = 0
for v in vecs {
tr = tr :: res[low,low+size(v)]
low = low + size(v)
}
tr
};
let transpose = fun(vecOfIndexable) {
let vecs = arrangedContiguously(vecOfIndexable)
let n = size(vecs[0]);
[[vecs[jx][ix] for jx in sequence(size(vecs))] for ix in sequence(n)]
};
let vectors = Vector.range(5000, {Vector.range(300)})
transpose(vectors)
""",
timeout = 30.0
)
self.assertTrue(result.isResult())
finally:
simulation.teardown()
def gbmRegressionFittingTest(self, nRows, nColumns, depth, nThreads, nBoosts, copies, report=True):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.dataGenerationScript(nRows, nColumns),
s3,
1,
timeout = 360,
memoryLimitMb = 30 * 1024,
threadCount = nThreads,
returnSimulation = True,
useInMemoryCache = False
)
try:
self.assertTrue(result.isResult())
dfPredictors, dfResponse = result.asResult.result
builder = simulation.compute(
self.regressionScript(depth, nBoosts),
timeout = 360,
dfResponse = dfResponse,
dfPredictors = dfPredictors
).asResult.result
t0 = time.time()
testName = self.getTestName(nRows, nColumns, depth, nBoosts, nThreads, copies)
result = simulation.compute(
"Vector.range(%s).apply(fun(x) { builder.fit(dfPredictors[,-x-1], dfResponse[,-x-1]) })"
% copies,
timeout = 360,
builder=builder,
dfPredictors=dfPredictors,
dfResponse=dfResponse,
).asResult.result
totalTimeToReturnResult = time.time() - t0
if report:
PerformanceTestReporter.recordTest(testName, totalTimeToReturnResult, None)
finally:
simulation.teardown()
def createSimulation(self,
useHdfsObjectStore=False,
objectStore=None,
sharedStateViewFactory=None,
workerCount=4,
machineIdHashSeed=None,
s3Service=None):
s3 = s3Service or InMemoryS3Interface.InMemoryS3InterfaceFactory()
return InMemoryCumulusSimulation.InMemoryCumulusSimulation(
workerCount,
1,
memoryPerWorkerMB=100,
threadsPerWorker=2,
s3Service=s3,
objectStore=objectStore,
sharedStateViewFactory=sharedStateViewFactory,
machineIdHashSeed=machineIdHashSeed)
def test_CalculationRicochet(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let f = fun(ct, seed = 1) {
let x = 0
let res = []
let it = iterator(math.random.UniformReal(0, size(v), seed))
for ix in sequence(ct) {
let x = Int64(pull it)
res = res :: (x / Float64(size(v)), v[x])
}
return res
}
v[2]
f(__count__,__seed__)
"""
vResult, sim = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
"Vector.range(125000000, math.log)",
s3,
4,
timeout = 120,
memoryLimitMb=400,
threadCount = 1,
useInMemoryCache = True,
returnSimulation = True
)
try:
v = vResult.asResult.result
t0 = time.time()
sim.compute(text.replace("__seed__", "1").replace("__count__", "1000"), timeout = 120, v = v)
PerformanceTestReporter.recordTest("python.InMemoryCumulus.Ricochet1000.Pass1", time.time() - t0,None)
t0 = time.time()
sim.compute(text.replace("__seed__", "2").replace("__count__", "1000"), timeout = 120, v = v)
PerformanceTestReporter.recordTest("python.InMemoryCumulus.Ricochet1000.Pass2", time.time() - t0,None)
finally:
sim.teardown()
def test_disk_scans(self):
s3 = ActualS3Interface.ActualS3InterfaceFactory()
objectStore = S3ObjectStore.S3ObjectStore(
s3, Setup.config().userDataS3Bucket, prefix="test_object_cache/")
_, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
"1+1",
s3,
1,
memoryLimitMb=1 * 1024,
threadCount=30,
returnSimulation=True,
ioTaskThreadOverride=8,
objectStore=objectStore,
useInMemoryCache=False #use an actual disk cache for this
)
try:
gigabytes = 8
t0 = time.time()
resultVectors = []
for ix in range(gigabytes):
result = simulation.compute("Vector.range(125000000 + %s)" %
ix,
timeout=120)
resultVectors.append(result.asResult.result)
t1 = time.time()
intResults = []
for vec in resultVectors:
result = simulation.compute("v.sum()", timeout=120, v=vec)
intResults.append(result.asResult.result.pyval)
self.assertTrue(len(intResults) == gigabytes)
PerformanceTestReporter.recordTest("python.BigBox.Disk.Write.10GB",
t1 - t0, None)
PerformanceTestReporter.recordTest(
"python.BigBox.Disk.WriteAndScan.%sGB" % gigabytes,
time.time() - t0, None)
finally:
simulation.teardown()
def compareCudaToCPU(self, funcExpr, vecExpr):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let f = __funcExpr__;
let i = __vecExpr__;
let cuda = `CUDAVectorApply(f, [i])[0];
let cpu = f(i)
if (cuda == cpu)
true
else
throw String(cuda) + " != " + String(cpu)
""".replace("__funcExpr__", funcExpr).replace("__vecExpr__", vecExpr)
res = InMemoryCumulusSimulation.computeUsingSeveralWorkers(text, s3, 1, timeout=120, threadCount=4)
self.assertIsNotNone(res)
self.assertTrue(res.isResult(), "Failed with %s on %s: %s" % (funcExpr, vecExpr, res))
def compareCudaToCPUnoCheck(self, funcExpr, vecExpr, captureExpr=""):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = captureExpr + """
let f = __funcExpr__;
let vec = __vecExpr__;
let cuda = cached`(#GpuApply(f, vec));
let cpu = [f(x) for x in vec]
if (cuda == cpu)
true
else
throw String(cuda) + " != " + String(cpu)
""".replace("__funcExpr__", funcExpr).replace("__vecExpr__", vecExpr)
res = InMemoryCumulusSimulation.computeUsingSeveralWorkers(text, s3, 1, timeout=120, threadCount=4)
self.assertIsNotNone(res)
return res
def regressionTreePredictionTest(self, mbOfData, columns, testName,
treeDepth, threads, minSamplesSplit=50):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.dataGenerationScript(mbOfData, columns),
s3,
1,
timeout = 360,
memoryLimitMb = 45 * 1024,
threadCount = threads,
returnSimulation = True,
useInMemoryCache = False
)
try:
self.assertTrue(result.isResult())
dfResponse, dfPredictors = result.asResult.result
fitTree = simulation.compute(
self.regressionScript(treeDepth, minSamplesSplit - 1),
timeout=120,
dfResponse=dfResponse,
dfPredictors=dfPredictors
)
def predictionScript(dirtyFlag=1):
return ";(%s; fitRegressionTree.predict(dfPredictors));" % dirtyFlag
t0 = time.time()
result = simulation.compute(
predictionScript(),
timeout=120,
dfPredictors=dfPredictors,
fitRegressionTree=fitTree.asResult.result
)
totalTimeToReturnResult = time.time() - t0
self.assertTrue(result.isResult())
PerformanceTestReporter.recordTest(testName, totalTimeToReturnResult, None)
finally:
simulation.teardown()
def test_takeFromLargeObjectsAsymmetric(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let N = 20;
//every thousandth string is 1 MB. Just take those.
let takeFrom = [
if (ix % 1000 == 0)
(" " * 100 * 100 * 10 * 10 + " " * (ix / 1000))
else
""
for ix in sequence(N * 1000)].paged;
let indices = Vector.range(N,fun(x) { x * 1000 }).paged;
let result = cached`(#ExternalIoTask(#DistributedDataOperation(#Take(indices, takeFrom))))
let targetResult = indices ~~ {takeFrom[_]};
assertions.assertEqual(size(result), size(targetResult))
assertions.assertEqual(result, targetResult)
result
"""
try:
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
text,
s3,
1,
timeout=TIMEOUT,
memoryLimitMb=1000,
returnSimulation = True,
pageSizeOverride = 1024 * 1024
)
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
for page in result.asResult.result.getVectorPageIds(simulation.getWorkerVdm(0)):
self.assertLess(page.bytecount / 1024.0 / 1024.0, 5.0)
finally:
simulation.teardown()
def test_multiboxDataTasksTake_2(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let N = 10 * 1000000;
let takeFrom = Vector.range(N)
let indices = Vector.range(N,fun(x) { (0, (x * 503) % N ) });
cached`(#ExternalIoTask(#DistributedDataOperation(#Take(indices, takeFrom))))[0]
"""
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
text,
s3,
8,
timeout=TIMEOUT,
memoryLimitMb=200,
returnSimulation=True)
logging.info("Simulation completed")
maxHighWatermark = 0
try:
for ix in range(8):
vdm = simulation.getWorkerVdm(ix)
vdmm = vdm.getMemoryManager()
logging.info("Total bytes: %s",
vdmm.getTotalBytesMmappedHighWaterMark())
maxHighWatermark = max(
maxHighWatermark, vdmm.getTotalBytesMmappedHighWaterMark())
vdm = None
vdmm = None
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
self.assertTrue(isinstance(result.asResult.result.pyval, int),
result)
finally:
simulation.teardown()
self.assertTrue(maxHighWatermark < 265 * 1024 * 1024)
def DISABLEDtest_canTriggerCheckpointOfCompleted(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
simulation = InMemoryCumulusSimulation.InMemoryCumulusSimulation(
4, #worker count
1,
memoryPerWorkerMB=100,
threadsPerWorker=2,
s3Service=s3)
try:
self.assertTrue(simulation.waitForGlobalScheduler(timeout=2.0))
simulation.waitForHandshake()
simulation.submitComputation("1+2")
result = simulation.waitForAnyResult()
self.assertTrue(result.isResult())
simulation.getGlobalScheduler(
).triggerFullCheckpointsOnOutstandingComputations()
ts = self.waitForFullCheckpoint(simulation, onlyUnfinished=False)
self.assertTrue(ts is not None)
statuses = simulation.getGlobalScheduler(
).currentOutstandingCheckpointStatuses(False, False)
status = statuses[0]
compId = status[0]
checkpointStatus = status[1][0]
checkpointRequest = status[1][1]
#verify that it's a storage checkpoint
self.assertTrue(checkpointRequest.writeToStorage)
self.assertTrue(checkpointStatus.checkpointSuccessful)
self.assertTrue(checkpointStatus.isRootComputationFinished)
finally:
simulation.teardown()
def test_multiboxDataTasksTake_2(self):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
text = """
let N = 10 * 1000000;
let takeFrom = Vector.range(N)
let indices = Vector.range(N,fun(x) { (0, (x * 503) % N ) });
cached`(#ExternalIoTask(#DistributedDataOperation(#Take(indices, takeFrom))))[0]
"""
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
text,
s3,
8,
timeout=TIMEOUT,
memoryLimitMb=200,
returnSimulation = True
)
logging.info("Simulation completed")
maxHighWatermark = 0
try:
for ix in range(8):
vdm = simulation.getWorkerVdm(ix)
vdmm = vdm.getMemoryManager()
logging.info("Total bytes: %s", vdmm.getTotalBytesMmappedHighWaterMark())
maxHighWatermark = max(maxHighWatermark, vdmm.getTotalBytesMmappedHighWaterMark())
vdm = None
vdmm = None
self.assertTrue(result is not None)
self.assertTrue(result.isResult(), result)
self.assertTrue(isinstance(result.asResult.result.pyval,int), result)
finally:
simulation.teardown()
self.assertTrue(maxHighWatermark < 265 * 1024 * 1024)
def computeUsingSeveralWorkers(self, *args, **kwds):
return InMemoryCumulusSimulation.computeUsingSeveralWorkers(*args, **kwds)
def dataframeSumTest(self, mbOfData, colCount, threadCount, recordResults = True):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
t0 = time.time()
randomColumnsToPick = 10
totalRowsToSum = 1000000
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.dataGenerationScript(mbOfData, colCount),
s3,
count=1,
timeout = 360,
memoryLimitMb = 10000,
threadCount = threadCount,
returnSimulation = True,
useInMemoryCache = False,
channelThroughputMBPerSecond = None
)
try:
self.assertTrue(result.isResult())
data = result.asResult.result
executionScript = ("""
let randomRowwiseSumFun = fun (row, randomColumnsToPick, baseSeed){
let rng = iterator(math.random.MultiplyWithCarry(baseSeed + row.rowIndex()));
let tr = nothing;
let ix = 0;
let rowSize = size(row)
while (ix < randomColumnsToPick) {
let nextIx = (pull rng) % rowSize;
tr = tr + row[nextIx]
ix = ix + 1
}
tr
}
let randomColumnsToPick = __subsetSize__;
let baseSeed = 5;
sum(0, __rows_to_sum__, fun(ix) { randomRowwiseSumFun(data[ix % size(data)], randomColumnsToPick, baseSeed) })
"""
.replace("__subsetSize__",str(randomColumnsToPick))
.replace("__rows_to_sum__",str(totalRowsToSum * threadCount))
)
t0 = time.time()
result = simulation.compute(
executionScript,
timeout=1080,
data=data
)
computeDuration = time.time() - t0
totalValuesAccessed = totalRowsToSum * randomColumnsToPick * threadCount
totalValuesPerSecondPerThread = totalValuesAccessed * 2 / computeDuration / threadCount
secondsToDo10MillionPerThread = 10 * 1000000 / totalValuesPerSecondPerThread
if recordResults:
PerformanceTestReporter.recordTest(
"python.BigBox.RandomColumnAccess.access10mm_%smb_%scols_%sthreads" % (
mbOfData,
colCount,
threadCount
),
secondsToDo10MillionPerThread,
None
)
self.assertTrue(result.isResult())
return computeDuration
finally:
dfResponse = None
dfPredictors = None
result = None
simulation.teardown()
def largeDatasetBigLMTest(self, mbOfData, columns, threads, machineCount, ratio = .5):
s3 = InMemoryS3Interface.InMemoryS3InterfaceFactory()
t0 = time.time()
result, simulation = InMemoryCumulusSimulation.computeUsingSeveralWorkers(
self.dataGenerationScript(mbOfData, columns),
s3,
machineCount,
timeout = 360,
memoryLimitMb = mbOfData / ratio / machineCount,
channelThroughputMBPerSecond = 100.0,
threadCount = threads,
returnSimulation = True,
useInMemoryCache = False
)
try:
self.assertTrue(result.isResult())
dfResponse, dfPredictors = result.asResult.result
regressionScript = """
let model = math.regression.LinearRegression(dfPredictors, dfResponse,coefficientsOnly:true);
let coefficients = model.coefficients();
coefficients[0]
"""
t0 = time.time()
result = simulation.compute(
regressionScript,
timeout=1080,
dfResponse=dfResponse,
dfPredictors=dfPredictors
)
totalTimeToReturnResult = time.time() - t0
self.assertTrue(result.isResult(), result)
self.assertTrue(result.isResult())
print "Done with the first regression"
regressionScript2 = """
let newCol = dfPredictors.rowApply(fun(row) { math.sin(row[0] ) })
let newCol2 = dfPredictors.rowApply(fun(row) { math.sin(row[0] + 1) })
let model2 = math.regression.LinearRegression(dfPredictors.addColumn(newCol).addColumn(newCol2), dfResponse, coefficientsOnly:true)
model2.coefficients()[0]
"""
result2 = simulation.compute(
regressionScript2,
timeout=1080,
dfResponse=dfResponse,
dfPredictors=dfPredictors
)
totalTimeToReturnResult = time.time() - t0
self.assertTrue(result2.isResult(), result2)
PerformanceTestReporter.recordTest(
"algorithms.linearRegression.inMemory_%sMB_%scols_%sthreads_%smachines" %
(mbOfData, columns,threads,machineCount),
totalTimeToReturnResult,
None
)
finally:
dfResponse = None
dfPredictors = None
result = None
simulation.teardown()