def test_run(self):
# part 1
imputer = MeanImputation(hyperparams=hp)
imputer.set_training_data(inputs=X)
imputer.fit(timeout=self.enough_time)
# print(imputer.get_params())
self.assertEqual(imputer._has_finished, True)
self.assertEqual(imputer._iterations_done, True)
result = imputer.produce(inputs=X, timeout=self.enough_time).value
self.helper_impute_result_check(X, result)
# part2: test set_params()
imputer2 = MeanImputation(hyperparams=hp)
imputer2.set_params(params=imputer.get_params())
self.assertEqual(imputer2._has_finished, True)
self.assertEqual(imputer2._iterations_done, True)
result2 = imputer2.produce(inputs=X, timeout=self.enough_time).value
self.assertEqual(result2.equals(result),
True) # two imputers' results should be same
self.assertEqual(imputer2._has_finished, True)
self.assertEqual(imputer2._iterations_done, True)
def test_noMV(self):
"""
test on the dataset has no missing values
"""
imputer = MeanImputation(hyperparams=hp)
imputer.set_training_data(inputs=X)
imputer.fit(timeout=self.enough_time)
result = imputer.produce(inputs=X, timeout=self.enough_time).value
# 1. check produce(): `result` contains no missing value
result2 = imputer.produce(inputs=result,
timeout=self.enough_time).value
self.assertEqual(result.equals(result2), True)
# 2. check fit() & get_params() try fit on no-missing-value dataset
imputer2 = MeanImputation(hyperparams=hp)
imputer2.set_training_data(inputs=result)
imputer2.fit(timeout=self.enough_time)
def test_notAlign(self):
"""
test the case that the missing value situations in trainset and testset are not aligned. eg:
`a` missing-value columns in trainset, `b` missing-value columns in testset.
`a` > `b`, or `a` < `b`
"""
imputer = MeanImputation(hyperparams=hp)
imputer.set_training_data(inputs=X)
imputer.fit(timeout=self.enough_time)
result = imputer.produce(inputs=X, timeout=self.enough_time).value
# PART1: when `a` > `b`
data2 = result.copy()
data2["T3"] = X["T3"].copy(
) # only set this column to original column, with missing vlaues
result2 = imputer.produce(inputs=data2, timeout=self.enough_time).value
self.helper_impute_result_check(data2, result2)
# PART2: when `a` < `b`
imputer = MeanImputation(hyperparams=hp)
imputer.set_training_data(inputs=data2)
imputer.fit(timeout=self.enough_time)
result = imputer.produce(inputs=X, timeout=self.enough_time).value
# data contains more missingvalue columns than data2,
# the imputer should triger default impute method for the column that not is trained
self.helper_impute_result_check(X, result)
# PART3: trunk the data : sample wise
imputer = MeanImputation(hyperparams=hp)
imputer.set_training_data(inputs=X)
imputer.fit(timeout=self.enough_time)
result = imputer.produce(inputs=X[0:20],
timeout=self.enough_time).value
self.helper_impute_result_check(X[0:20], result)
print(trainData.head())
print(trainTargets.head())
print(np.asarray(trainTargets['Class']))
print(testData.head())
hp = EncHyperparameter.sample()
enc = Encoder(hyperparams=hp)
enc.set_training_data(inputs=trainData)
enc.fit()
encodedData = enc.produce(inputs=trainData).value
encodedTestData = enc.produce(inputs=testData).value
# Initialize the DSBox imputer
hp = MeanHyperparameter.sample()
imputer = MeanImputation(hyperparams=hp)
imputer.set_training_data(inputs=encodedData) # unsupervised
imputer.fit(timeout=100) # give 100 seconds to fit
print("\nParams:")
print(imputer.get_params())
imputer2 = MeanImputation(hyperparams=hp)
imputer2.set_params(params=imputer.get_params())
imputedData = imputer2.produce(inputs=encodedData, timeout=100).value
model = BaggingClassifier()
trainedModel = model.fit(imputedData, np.asarray(trainTargets['Class']))
predictedTargets = trainedModel.predict(
imputer.produce(inputs=encodedTestData).value)
def text2int(col):
"""
convert column value from text to integer codes (0,1,2...)
"""
return pd.DataFrame(col.astype('category').cat.codes, columns=[col.name])
import pandas as pd
from dsbox.datapreprocessing.cleaner import MeanImputation
# STEP 1: get data
data_path = "../../dsbox-data/o_38/encoded/"
data_name = data_path + "trainData_encoded.csv"
label_name = data_path + "trainTargets_encoded.csv" # make sure your label target is in the second column of this file
data = pd.read_csv(data_name)
label = text2int(pd.read_csv(label_name)["Class"])
data.drop("d3mIndex", axis=1) # drop because id, useless
# STEP 2: go to use the Imputer !
imputer = MeanImputation()
imputer.set_params(verbose=1)
imputer.set_training_data(inputs=data) # unsupervised
imputer.fit(timeout=10) # give 10 seconds to fit
print(imputer.get_call_metadata()) # to see wether fit worked
result = imputer.produce(inputs=data, timeout=10)
print(imputer.get_call_metadata()) # to see wether produce worked