def crossfolding(trainrecipes):
logging.info("### run summary ###")
#initialize counters (at least 2)
numPartitions = 10
accuracy = np.array([0.0] * numPartitions)
partitionsSize = np.array([0] * numPartitions)
#define partitions size
if len(trainrecipes) < numPartitions:
logging.error("Train dataset must have more than %d items" %
numPartitions)
sys.exit(0)
partitionsSize += len(trainrecipes) / numPartitions
for i in range(len(trainrecipes) % numPartitions):
partitionsSize[i] += 1
logging.info(">number of training recipes: %d" % len(trainrecipes))
#calculate accuracy for each partition
logging.info("...calculating accuracy for each partition...")
partitionIndex = 0
for i in range(numPartitions):
logging.info("FOLD %d" % (i + 1))
#get train and test lists
testList = trainrecipes[partitionIndex:partitionIndex +
partitionsSize[i]]
trainList = [] * (len(trainrecipes) - len(testList))
for nDocument in range(len(trainrecipes)):
if (nDocument < partitionIndex) | (
nDocument > partitionIndex + partitionsSize[i]):
trainList.append(trainrecipes[nDocument])
partitionIndex += partitionsSize[i]
#classify test list
classifiedList = bayes.run(trainList, testList)
totalrecipes = 0.0
truePositives = 0.0
for recipe in testList:
totalrecipes += 1
if classifiedList[recipe['id']] == recipe['cuisine']:
truePositives += 1
#compare classification to calculate accuracy
accuracy[i] = truePositives / totalrecipes
#calculate avg accuracy
avgAccuracy = 0.0
avgAccuracy = np.average(accuracy)
return avgAccuracy
def main(argv):
#load json files.
with open(args.trainRecipesFile) as train_recipes_file:
trainJson = json.load(train_recipes_file)
with open(args.unknownRecipesFile) as unknown_recipes_file:
unknownJson = json.load(unknown_recipes_file)
results = {}
#run naive bayes classifier.
results = bayes.run(trainJson,unknownJson)
# write to output file
text_file = open(args.outputFile, "w")
text_file.write('id,cuisine\n')
for i in results:
text_file.write(str(i) + ',' + results[i] + '\n')
text_file.close()
def main(argv):
#load json files.
with open(args.trainRecipesFile) as train_recipes_file:
trainJson = json.load(train_recipes_file)
with open(args.unknownRecipesFile) as unknown_recipes_file:
unknownJson = json.load(unknown_recipes_file)
results = {}
#run naive bayes classifier.
results = bayes.run(trainJson, unknownJson)
# write to output file
text_file = open(args.outputFile, "w")
text_file.write('id,cuisine\n')
for i in results:
text_file.write(str(i) + ',' + results[i] + '\n')
text_file.close()
def crossfolding(trainrecipes):
logging.info("### run summary ###")
#initialize counters (at least 2)
numPartitions = 10
accuracy = np.array([0.0] * numPartitions)
partitionsSize = np.array([0] * numPartitions)
#define partitions size
if len(trainrecipes) < numPartitions:
logging.error("Train dataset must have more than %d items" % numPartitions)
sys.exit(0)
partitionsSize += len(trainrecipes) / numPartitions
for i in range(len(trainrecipes) % numPartitions):
partitionsSize[i] += 1
logging.info(">number of training recipes: %d" % len(trainrecipes))
#calculate accuracy for each partition
logging.info("...calculating accuracy for each partition...")
partitionIndex = 0
for i in range(numPartitions):
logging.info("FOLD %d" % (i+1))
#get train and test lists
testList = trainrecipes[partitionIndex:partitionIndex+partitionsSize[i]]
trainList = [] * (len(trainrecipes)-len(testList))
for nDocument in range(len(trainrecipes)):
if (nDocument < partitionIndex) | (nDocument>partitionIndex+partitionsSize[i]):
trainList.append(trainrecipes[nDocument])
partitionIndex += partitionsSize[i]
#classify test list
classifiedList = bayes.run(trainList,testList)
totalrecipes = 0.0
truePositives = 0.0
for recipe in testList:
totalrecipes += 1
if classifiedList[recipe['id']] == recipe['cuisine']:
truePositives += 1
#compare classification to calculate accuracy
accuracy[i] = truePositives / totalrecipes
#calculate avg accuracy
avgAccuracy = 0.0
avgAccuracy = np.average(accuracy)
return avgAccuracy