def test_fillna_numerical():
df_test = InsolverDataFrame(pd.DataFrame(data={'col1': [1, 2, np.nan]}))
df_transformed = InsolverTransform(df_test, [
AutoFillNATransforms(),
])
df_transformed.ins_transform()
assert df_transformed['col1'][2] == 1.5
def predict():
# Request logging
current_datatime = strftime('[%Y-%b-%d %H:%M:%S]')
ip_address = request.headers.get("X-Forwarded-For", request.remote_addr)
logger.info(f'{current_datatime} request from {ip_address}: {request.json}')
start_prediction = time()
json_input = request.json
json_str = json.dumps(json_input['df'])
df = pd.read_json(json_str)
InsDataFrame = InsolverDataFrame(df)
# Apply transformations
InsTransforms = InsolverTransform(InsDataFrame, tranforms)
InsTransforms.ins_transform()
# Prediction
predicted = model.predict(InsTransforms)
result = {
'predicted': predicted.tolist()
}
# Response logging
end_prediction = time()
duration = round(end_prediction - start_prediction, 6)
current_datatime = strftime('[%Y-%b-%d %H:%M:%S]')
logger.info(f'{current_datatime} predicted for {duration} msec: {result}\n')
return jsonify(result)
def test_fillna_categorical():
df_test = InsolverDataFrame(
pd.DataFrame(data={'col1': ['A', 'B', 'C', 'A', None]}))
df_transformed = InsolverTransform(df_test, [
AutoFillNATransforms(),
])
df_transformed.ins_transform()
assert df_transformed['col1'][4] == 'A'
def test_fillna_categorical_all_na():
df_test = InsolverDataFrame(
pd.DataFrame(data={'col1': [None, None, None]}))
df_transformed = InsolverTransform(df_test, [
AutoFillNATransforms(),
])
df_transformed.ins_transform()
assert df_transformed['col1'][0] == 1
assert df_transformed['col1'][1] == 1
assert df_transformed['col1'][2] == 1
def test_EncoderTransforms():
df_test = InsolverDataFrame(
pd.DataFrame(data={'col1': ['A', 'B', 'C', 'A']}))
df_transformed = InsolverTransform(df_test, [
EncoderTransforms(['col1']),
])
df_transformed.ins_transform()
assert df_transformed['col1'][0] == 0
assert df_transformed['col1'][1] == 1
assert df_transformed['col1'][2] == 2
assert df_transformed['col1'][3] == 0
def test_OneHotEncoderTransforms():
df_test = InsolverDataFrame(
pd.DataFrame(data={'col1': ['A', 'B', 'C', 'A']}))
df_transformed = InsolverTransform(df_test, [
OneHotEncoderTransforms(['col1']),
])
df_transformed.ins_transform()
assert 'col1_A' in df_transformed.columns
assert 'col1_B' in df_transformed.columns
assert 'col1_C' in df_transformed.columns
assert df_transformed['col1_A'][0] == 1
assert df_transformed['col1_B'][1] == 1
assert df_transformed['col1_C'][2] == 1
assert df_transformed['col1_A'][3] == 1
async def predict(data: Data):
# Extract data in correct order
data_dict = data.dict()
df = pd.DataFrame(data_dict['df'])
InsDataFrame = InsolverDataFrame(df)
# Apply transformations
InsTransforms = InsolverTransform(InsDataFrame, tranforms)
InsTransforms.ins_transform()
# Prediction
predicted = model.predict(InsTransforms)
result = {'predicted': predicted.tolist()}
return result
df.dropna(how='all', axis=1, inplace=True)
df = df[df.ClaimAmount > 0]
InsDataFrame = InsolverDataFrame(df)
InsTransforms = InsolverTransform(InsDataFrame, [
TransformAge('DrivAge', 18, 75),
TransformExp('LicAge', 57),
TransformMapValues('Gender', {
'Male': 0,
'Female': 1
}),
TransformMapValues('MariStat', {
'Other': 0,
'Alone': 1
}),
TransformAgeGender('DrivAge',
'Gender',
'Age_m',
'Age_f',
age_default=18,
gender_male=0,
gender_female=1),
TransformPolynomizer('Age_m'),
TransformPolynomizer('Age_f'),
])
InsTransforms.ins_transform()
InsTransforms.save('transforms.pkl')
train, valid, test = InsTransforms.split_frame(val_size=0.15,
import pickle
import pandas as pd
from insolver import InsolverDataFrame
from insolver.transforms import InsolverTransform, init_transforms
from insolver.wrappers import InsolverGLMWrapper
# load data
df = pd.read_json('request_example.json')
InsDataFrame = InsolverDataFrame(df)
# load and init transformations
with open('transforms.pkl', 'rb') as file:
transforms = pickle.load(file)
transforms = init_transforms(transforms, inference=True)
# Apply transformations
InsTransforms = InsolverTransform(InsDataFrame, transforms)
InsTransforms.ins_transform()
# Load saved model
new_iglm = InsolverGLMWrapper(backend='h2o',
load_path='./insolver_glm_h2o_1605026853331')
predict_glm = new_iglm.predict(df)
print(predict_glm)