def test_predict_explain(self, binary_problem=False):
__labeled_inp_df = self.labeled_inp_df.copy(deep=True)
__nonlabeled_inp_df = self.nonlabeled_inp_df.copy(deep=True)
__labeled_binary_inp_df = self.labeled_binary_inp_df.copy(deep=True)
__schema_with_label = self.schema_with_label.copy()
ac = Classifier(model_configuration=testModelConfiguration)
if binary_problem:
ac.train(input_df=__labeled_binary_inp_df,
schema=__schema_with_label)
else:
ac.train(input_df=__labeled_inp_df, schema=__schema_with_label)
lr, fm, lm = ac.get_models()
ac = Classifier(model_configuration=testModelConfiguration)
ac.load_models(lr, fm, lm)
res_df = ac.predict_explain(input_df=__nonlabeled_inp_df)
self.assertTrue(isinstance(res_df, pd.DataFrame))
self.assertEqual(len(res_df.columns),
len(self.fields_without_label) + 4)
self.assertEqual(res_df.dtypes[-1], "object")
self.assertEqual(res_df.dtypes[-2], "float64")
self.assertEqual(res_df.dtypes[-3], "object")
self.assertEqual(res_df.dtypes[-4], "object")
#Test if all top-contributed features are present
def chk_contributor_existance(row):
contributors = row["TopContributors"].split(';')
features = [contrib.split('=')[0] for contrib in contributors]
for feat in features:
if '::' in feat:
field_name, field_value = feat.split('::')
self.assertIn(field_name, self.fields_without_label)
fld_no = list(res_df.columns).index(field_name)
if self.schema_without_label[fld_no] in ["text", "set"]:
#self.assertIn(field_value, row[field_name].lower())
self.assertTrue(True)
elif self.schema_without_label[fld_no] in [
"string", "numeric", "boolean"
]:
self.assertEqual(field_value, row[field_name])
else:
field_name = feat
if len(field_name) > 0:
self.assertIn(field_name, self.fields_without_label)
res_df.apply(chk_contributor_existance, axis=1)
def classify(cachedModelID, data):
startedTime = datetime.datetime.now()
assert(cachedModelID in cachedMSR), "Model not found."
model = cachedMSR[cachedModelID]['selectedModel']
emptyResults = {
'id': -1,
'classSummaries': []
}
#debug
print('Received a dataset with ', len(data['features']), ' features to classify.')
if (len(data['features']) ==0):
print('There is no feature, empty result set is returned.')
return emptyResults
print('Received a dataset with ', len(data['features'][0]['data']), ' rows to classify.')
if (len(data['features'][0]['data']) ==0):
print('There is no data, empty result set is returned.')
return emptyResults
candidate = model["candidate"]
features = candidate["features"]
config = candidate["config"]
unlabeled_df = datasetToDataframe(data)
filtered_input_df = unlabeled_df.filter([f['name'] for f in features])
lr, fm, lm = loadTrainedModel(model)
ac = Classifier(model_configuration=config)
ac.load_models(lr, fm, lm)
res_df = ac.predict_explain(input_df=filtered_input_df, topN_features=10)
reccom_df = ac.input_qlty(input_df=filtered_input_df, topN=10)
res_df = pd.concat([res_df, reccom_df.filter(["SuggestedFeatures"])], axis=1)
plCountSeries = res_df.groupby('PredictedLabel').PredictedLabel.count()
labels = list(plCountSeries.keys())
classSummaries = []
for label in labels:
filtered_res_df = res_df[res_df.PredictedLabel == label]
entropies = []
probabilities = []
results = []
for data_index, row in filtered_res_df.iterrows():
entropies.append(float(row.Entropy))
probsDict, allLabels = unpackProbs(row.Probabilities)
probabilities.append(float(probsDict[label]))
contributors = unpackContribs(row.TopContributors)
recommends = unpackSuggestedFeatures(row.SuggestedFeatures)
input_data = []
for feat in data['features']:
input_data.append({'id': id(), 'feature': feat['feature'], 'data': [feat['data'][data_index]]})
data_instance = {
'id': id(),
'dataset': { 'id': id(),
'features': input_data},
'index': data_index
}
classificationResult = {
'id': id(),
'allLabels': allLabels,
'entropy': float(row.Entropy),
'contributors': contributors,
'dataInstance': data_instance,
'predictedLabel': {
'id': id(),
'label': label,
'probability': float(probsDict[label])
},
'recommends': recommends
}
results.append(classificationResult)
classSumary = {
'id': id(),
'label': label,
'numInstances': int(plCountSeries[label]),
'probabilities': probabilities,
'entropies': entropies,
'results': results
}
classSummaries.append(classSumary)
batchClassificationResult = {
'id': id(),
"classSummaries": classSummaries
}
print('Classification time:' + str((datetime.datetime.now() - startedTime).total_seconds()) + ' seconds ')
return batchClassificationResult