def main():
# constructs the object car_insurance_model by loading in the data to the class DataModeler
car_insurance_model = DataModeler(
pd.read_csv("Data_In/Car_Insurance/DS_Assessment.csv"),
pd.DataFrame({'null_df': []}))
####################################################################################################################
# EXPLORATION
####################################################################################################################
# Get a very broad understanding of the data from the dimension and first 5 rows
# Prints the dimension of the data after successfully storing the data set as a Pandas data frame
print("The dimension of the car insurance data is: ",
car_insurance_model._train_data_set.shape)
print(car_insurance_model._train_data_set.head()
) # prints the first 5 rows of the data set
####################################################################################################################
# Example of some of the graphs used to explore the data for the attribute: Age
# These methods can and were used for all the attributes
'''
# counts number of each age
car_insurance_model.attribute_value_count('Age')
# counts Sale or NoSale for each number of age
car_insurance_model.attribute_value_count_by_classification('Age', 'Sale')
# displays and saves a bar graph showing the percentage of each value for the attribute Age in the data set
car_insurance_model.bar_graph_attribute('Age')
# displays a stacked Sale and NoSale bar graph for each attribute in Age in the data set
car_insurance_model.bar_graph_attribute_by_classification('Age', 'Sale')
# prints a summary of the distribution of the column 'Age' such as mean, standard deviation etc
car_insurance_model.describe_attribute('Age')
# plots a histogram of the attribute Age and also a quantile quantile plot
# car_insurance_model.histogram_and_q_q('Age')
# plots a scatter plot of Age and Price
car_insurance_model.scatter_plot('Age', 'Price')
# plots a scatter plot of Age and Price for Sale and NoSale
car_insurance_model.scatter_plot_by_classification('Age', 'Price', 'Sale')
car_insurance_model.histogram_and_q_q('Credit_Score')
####################################################################################################################
# Observe how much data is missing for each attribute
car_insurance_model.train_missing_data_ratio_print()
# displays and saves a bar graph of the percentage of missing values
car_insurance_model.missing_data_ratio_bar_graph()
car_insurance_model.heat_map()
'''
####################################################################################################################
# PROCESSING
####################################################################################################################
# Attempted to log and sqrt transform some skewed parameters however, I found the models to perform worse hence I
# decided to instead normalise the attributes to have a mean of 0 and standard deviation of 1. Code below
# demonstrates some of my attempts to better fit the data to a normal distribution
'''
car_insurance_model.histogram_and_q_q('Credit_Score')
# max_price = car_insurance_model._data_set['Price'].max()
# car_insurance_model._data_set['Price'] = max_price + 1 - car_insurance_model._data_set['Price']
car_insurance_model.boxcox_trans_attribute('Credit_Score', 0.1)
# car_insurance_model._data_set['Price'] = np.sqrt(car_insurance_model._data_set['Price'])
car_insurance_model.histogram_and_q_q('Credit_Score')
'''
####################################################################################################################
# Normalise attributes to a mean of zero and standard deviation of 1 before imputing
attributes_to_normalise = [
'Veh_Mileage', 'Credit_Score', 'License_Length', 'Veh_Value', 'Price',
'Age', 'Tax'
]
for i in attributes_to_normalise:
car_insurance_model.normalise_attribute(i)
####################################################################################################################
# creating new features from the attribute date
# decided to add day_of_the_week column to see if any information can be extracted
####################################################################################################################
# convert type of column date form object to datetime64
car_insurance_model._train_data_set['Date'] = pd.to_datetime(
car_insurance_model._train_data_set['Date'],
infer_datetime_format=True)
# add new column named days_of_the_week that has the day of the week
car_insurance_model._train_data_set = car_insurance_model._train_data_set.assign(
days_of_the_week=car_insurance_model._train_data_set['Date'].dt.
weekday_name)
####################################################################################################################
# bar graph of new column to see if any new information can be obtained
# car_insurance_model.bar_graph_attribute_by_classification('days_of_the_week', 'Sale')
# can see that on Friday typically there are less sales hence decided to create new column
# used similar method to extract month and year, found month would have added too many attributes when one hot
# encoding and year to not have any significant difference between 2015 and 2016
# one hot encodes the column days_of_the_week by adding 7 new attributes
car_insurance_model.one_hot_encode_attribute('days_of_the_week')
# drop date as there are so many different days
car_insurance_model.drop_attribute('Date')
####################################################################################################################
# Dealing with the attributes Tax and Price
# scatter plot the two attributes as they appear very highly correlated and could be used to impute the data
# car_insurance_model.scatter_plot_by_classification("Tax", "Price")
# found that tax and price follow two linear equations using car_insurance_model.scatter_plot("Tax", "Price")
# the cutoff between following either equation was when the tax was between a value of 32 to 35 which was found by
# looking through the data set:
# typically when tax < 34, tax = 0.05 * price and when tax > 34, tax = 0.1 * price
# hence this can be used to impute missing values more accurately
# compare how many values are imputed using this method
car_insurance_model.train_missing_data_ratio_print()
####################################################################################################################
# impute price
# as only 5 values are missing for both Price and Tax, the mean is imputed for these values
# create a list of the index of the missing values in the price attribute
price_missing_value_index = car_insurance_model._train_data_set[
car_insurance_model._train_data_set['Price'].isnull()].index.tolist()
# loop through the missing value index for price
for i in price_missing_value_index:
# if the value for tax in the same column as price is greater than 34
if car_insurance_model._train_data_set['Tax'].values[i] > 33:
# set the value for price equal to ten times the value of tax in the same column
car_insurance_model._train_data_set['Price'].values[
i] = car_insurance_model._train_data_set['Tax'].values[i] * 10
else:
# else set the price to 5 times the tax
car_insurance_model._train_data_set['Price'].values[
i] = car_insurance_model._train_data_set['Tax'].values[i] * 5
print('The number of price values imputed is ',
len(price_missing_value_index))
####################################################################################################################
# impute tax
# create a list of the index of the missing values in the tax attribute
tax_missing_value_index = car_insurance_model._train_data_set[car_insurance_model._train_data_set['Tax'].isnull()].\
index.tolist()
# loop through the missing value index for price
for i in tax_missing_value_index:
# if the value for tax in the same column as price is greater than 34
if car_insurance_model._train_data_set['Price'].values[i] > 330:
# set the value for price equal to ten times the value of tax in the same column
car_insurance_model._train_data_set['Tax'].values[
i] = car_insurance_model._train_data_set['Price'].values[
i] * 0.1
else:
# else set the price to 5 times the tax
car_insurance_model._train_data_set['Tax'].values[
i] = car_insurance_model._train_data_set['Price'].values[
i] * 0.05
print('The number of tax values imputed is ', len(tax_missing_value_index))
####################################################################################################################
car_insurance_model.train_missing_data_ratio_print()
# as only 5 values are missing for both Price and Tax, the mean is imputed for these values
car_insurance_model.impute_mean('Price')
car_insurance_model.impute_mean('Tax')
####################################################################################################################
# one hot encoding certain attributes
car_insurance_model.one_hot_encode_attribute(
'Marital_Status') # one hot encodes Marital_Status
####################################################################################################################
# found credit score to have an interesting value of 9999 for some customers, I attempted to one hot encode all the
# customers that had this score to a new column however, found this to have no significant difference on the model
# however, I decided to leave the code in the class DataPreprocessor:
# create a list of the indices of the credit score with a score of 9999
credit_score_9999_index = car_insurance_model._train_data_set[
car_insurance_model._train_data_set['Credit_Score'] ==
9999].index.tolist()
# create a new column for the credit scores with 9999 so they can be put into a different attribute
car_insurance_model._train_data_set['Infinite_Credit_Score'] = 0
for i in credit_score_9999_index:
# set the new column values to 1 (one hot encoding)
car_insurance_model._train_data_set['Infinite_Credit_Score'].values[
i] = 1
# drop the credit score of 9999 from the attribute Credit_Score
car_insurance_model._train_data_set['Credit_Score'].values[i] = 0
####################################################################################################################
# attempted to impute using knn from a package known as fancyimpute however, I found this to be extremely
# inefficient and instead used standard methods. the code is left the class DataPreprocessor and called on the next
# line:
# car_insurance_model.impute_knn(3)
####################################################################################################################
# Impute the other attributes using standard methods
car_insurance_model.impute_median('Credit_Score')
car_insurance_model.impute_mode('Veh_Mileage')
car_insurance_model.impute_median(
'License_Length'
) # should try to impute by first categorising by Maritial_Status
car_insurance_model.impute_mode('Veh_Value') # should use a better method
car_insurance_model.impute_median('Age')
####################################################################################################################
# check all values have been imputed
print(
'After imputing all the attributes, the missing ratio is found to be:')
car_insurance_model.train_missing_data_ratio_print()
####################################################################################################################
# MODELS
####################################################################################################################
car_insurance_model.shuffle_data_set(
) # shuffle the data set before splitting
# split data to 75% training, 25% test with a seed set to 2 (to get the same split when running the code
car_insurance_model.split_data_set_if_test_not_split('Sale', 0.25, 2)
####################################################################################################################
# Knn model
# gridsearch for knn
# uncomment to run grid search
grid_parameters_knn = [{'n_neighbors': [5, 15, 19]}]
car_insurance_model.classification_model_grid_search(
KNeighborsClassifier, grid_parameters_knn, 2)
# fit a knn with k=5 and print percentage accuracy for 10-fold cross validation and confusion matrix against the
# test set
tuned_parameters_knn = {'n_neighbors': 19}
car_insurance_model.classification_model(KNeighborsClassifier,
tuned_parameters_knn, 10)
####################################################################################################################
# SMV model
# found these set of parameters to be the most optimum when performing a grid search
# uncomment to run grid search
tuned_parameters_svm = [{
'kernel': ['rbf'],
'gamma': [1 / 15, 1 / 16, 1 / 17],
'C': [11, 10, 12],
'decision_function_shape': ['ovo']
}]
car_insurance_model.classification_model_grid_search(
SVC, tuned_parameters_svm, 2)
# fit a svm and print percentage accuracy for 10-fold cross and shows the confusion matrix for the best
# hyper-parameters found when performing the grid-search
# k-fold cross validation for optimum hyper-parameters to validate SVM model
tuned_parameters_svm = {
'C': 10,
'decision_function_shape': 'ovo',
'degree': 3,
'gamma': 1 / 16,
'kernel': 'rbf'
}
car_insurance_model.classification_model(SVC, tuned_parameters_svm, 10)
def main():
####################################################################################################################
# Main used for house price data set
# at this point once the data has been explored, want train_Y to be in its own variable separate from train_X to
# pre-process the data train_X and test_X should not be combined at any point as the data should be preprocessed in
# one go for train_X but in a real world scenario, test_X may not come in as a large dataset
model_house = DataModeler(pd.read_csv("Data_In/House_Prices/train.csv"),
pd.read_csv("Data_In/House_Prices/test.csv"))
# model_house.box_plot('SalePrice', 'YearBuilt')
# model_house.bar_graph_attribute('YrSold')
# model_house.line_graph_percentage_difference('YrSold')
# dropped after looking at a scatter plot of the two attributes
# model_house.scatter_plot('SalePrice', 'GrLivArea')
model_house.drop_outliers_target_less_y_attribute_greater_x(
'SalePrice', 300000, 'GrLivArea', 4000)
model_house.drop_outliers_target_greater_y_attribute_greater_x(
'SalePrice', 200000, 'LotFrontage', 300)
# model_house.scatter_plot('SalePrice', 'GrLivArea')
print(model_house._train_data_set.shape)
model_house.switch_na_to_median_other_attribute('LotFrontage',
'Neighborhood')
# all features in train are all pub and 2 na, 'NoSewa' is in test set hence the attribute doesnt help in any way
# with the model so it is dropped
attributes_to_none = [
"PoolQC", "MiscFeature", "Alley", "Fence", "FireplaceQu", "GarageCond",
"GarageQual", "GarageFinish", "GarageYrBlt", "GarageType",
"BsmtFinType2", "BsmtExposure", "BsmtFinType1", "BsmtCond", "BsmtQual",
"MasVnrType"
]
attributes_to_zero = [
'MasVnrArea', 'BsmtHalfBath', 'BsmtFullBath', 'GarageArea',
'GarageCars', 'TotalBsmtSF', 'BsmtUnfSF', 'BsmtFinSF2', 'BsmtFinSF1'
]
attributes_to_mode = [
'Electrical', 'MSZoning', 'Functional', 'SaleType', 'KitchenQual',
'Exterior2nd', 'Exterior1st'
]
attributes_to_categorical = [
'MSSubClass', 'OverallCond', 'YrSold', 'MoSold'
]
for x in attributes_to_none:
model_house.impute_none(x)
for x in attributes_to_zero:
model_house.impute_zero(x)
for x in attributes_to_mode:
model_house.impute_mode(x)
for x in attributes_to_categorical:
model_house.convert_attribute_to_categorical(x)
# drops id column from train_X and test_X to move it to _test_y_id
model_house.index_column_drop_and_move_to_pred_y('Id')
model_house.move_target_to_train_y(
'SalePrice') # moves saleprice to train_Y
model_house.train_missing_data_ratio_print()
model_house.test_missing_data_ratio_print()
####################################################################################################################
# all the missing values are inputted!!!!
####################################################################################################################
# print('The dimension of the train is', model_house._train_data_set.shape)
# print('The dimension of the test is', model_house._test_data_set.shape)
####################################################################################################################
# need to add these in later
attributes_to_drop = [
'Utilities', 'YearBuilt', 'MoSold', 'YrSold', 'GarageYrBlt',
'YearRemodAdd'
]
for x in attributes_to_drop:
model_house.drop_attribute(x)
####################################################################################################################
attributes_to_normalise = [
'LotFrontage', 'LotArea', 'MasVnrArea', 'BsmtFinSF1', 'BsmtFinSF2',
'BsmtUnfSF', 'TotalBsmtSF', '1stFlrSF', '2ndFlrSF', 'LowQualFinSF',
'GrLivArea', 'GarageArea', 'WoodDeckSF', 'OpenPorchSF',
'EnclosedPorch', '3SsnPorch', 'ScreenPorch', 'PoolArea', 'MiscVal'
]
for x in attributes_to_normalise:
model_house.normalise_attribute(x)
attributes_to_one_hot_encode = [
'MSSubClass', 'MSZoning', 'Street', 'Alley', 'LotShape', 'LandContour',
'LotConfig', 'LandSlope', 'Neighborhood', 'Condition1', 'Condition2',
'BldgType', 'HouseStyle', 'RoofStyle', 'RoofMatl', 'Exterior1st',
'Exterior2nd', 'MasVnrType', 'ExterQual', 'ExterCond', 'Foundation',
'BsmtQual', 'BsmtCond', 'BsmtExposure', 'BsmtFinType1', 'BsmtFinType2',
'Heating', 'HeatingQC', 'CentralAir', 'Electrical', 'KitchenQual',
'Functional', 'GarageType', 'FireplaceQu', 'GarageFinish',
'GarageQual', 'GarageCond', 'PavedDrive', 'PoolQC', 'Fence',
'MiscFeature', 'SaleType', 'OverallCond', 'SaleCondition',
'OverallQual', 'FullBath', 'BedroomAbvGr', 'TotRmsAbvGrd',
'GarageCars', 'BsmtFullBath', 'KitchenAbvGr', 'BsmtHalfBath',
'HalfBath', 'Fireplaces'
]
for x in attributes_to_one_hot_encode:
model_house.one_hot_encode_attribute(x)
model_house.delete_unnecessary_one_hot_encoded_columns()
# print('The dimension of the train is', model_house._train_data_set.shape)
# print('The dimension of the test is', model_house._test_data_set.shape)
# could transform target before and after
# model_house.box_cox_target(0.1)
####################################################################################################################
# Lasso
model_house.lasso_compare_alpha([75, 100, 200, 300]).to_csv(
'Data_Out/Lasso_model_alpha_1_0point1_0point01.csv', index=False)
lasso_model_grid_parameters = [{'alpha': [75, 100, 200, 300]}]
model_house.regression_model_grid_search(Lasso,
lasso_model_grid_parameters, 10)
lasso_model_tuned_parameters = {'alpha': 1000, 'random_state': 1}
model_house.regression_model_submission(Lasso, 'SalePrice',
lasso_model_tuned_parameters)
####################################################################################################################
# ridge regression optimised
ridge_model_grid_parameters = [{'alpha': [1, 5, 7, 10]}]
model_house.regression_model_grid_search(Ridge,
ridge_model_grid_parameters, 10)
ridge_model_tuned_parameters = {'alpha': 10.0}
model_house.regression_model_submission(Ridge, 'SalePrice',
ridge_model_tuned_parameters)
####################################################################################################################
# kernel ridge regression gridsearch
kernel_ridge_model_grid_parameters = {
'alpha': [5, 9, 10, 11],
'kernel': ['linear'],
'degree': [1, 2, 3]
}
model_house.regression_model_grid_search(
KernelRidge, kernel_ridge_model_grid_parameters, 10)
kernel_ridge_model_fine_tuned_parameters = {
'alpha': 9,
'kernel': 'linear',
'degree': 1
}
model_house.regression_model_submission(
KernelRidge, 'SalePrice', kernel_ridge_model_fine_tuned_parameters)
####################################################################################################################
# linear optimised
linear_model_grid_parameters = {
'fit_intercept': [True, False],
'normalize': [True, False],
'copy_X': [True, False]
}
model_house.regression_model_grid_search(LinearRegression,
linear_model_grid_parameters, 10)
linear_model_fine_tuned_parameters = {
'fit_intercept': True,
'normalize': True,
'copy_X': False
}
model_house.regression_model_submission(
LinearRegression, 'SalePrice', linear_model_fine_tuned_parameters)
def main():
####################################################################################################################
# Main used for house price data set
# at this point once the data has been explored, want train_Y to be in its own variable separate from train_X to
# pre-process the data train_X and test_X should not be combined at any point as the data should be preprocessed in
# one go for train_X but in a real world scenario, test_X may not come in as a large dataset
model_adult = DataModeler(pd.read_csv("Data_In/Adult/adult.data.txt", header=None, sep=",\s", na_values=["?"])
, pd.read_csv("Data_In/Adult/adult.test.txt", header = None, sep=",\s", na_values=["?"]))
model_adult._train_data_set.columns = ["age", "workclass", "fnlwgt", "education", "education-num", "marital-status",
"occupation", "relationship", "race", "sex", "capital-gain", "capital-loss",
"hours-per-week", "native-country", "salary"]
print(model_adult._train_data_set)
model_adult._test_data_set.columns = ["age", "workclass", "fnlwgt", "education", "education-num", "marital-status",
"occupation", "relationship", "race", "sex", "capital-gain", "capital-loss",
"hours-per-week", "native-country", "salary"]
#model_adult.box_plot("age","salary")
#model_adult.missing_data_ratio_bar_graph()
#model_adult.scatter_plot("hours-per-week","age")
#model_adult.train_missing_data_ratio_print()
# model_adult.histogram_and_q_q("age")
# model_adult.histogram_and_q_q("hours-per-week")
# model_adult.histogram_and_q_q("capital-gain")
# model_adult.histogram_and_q_q("capital-loss")
# # model_adult.bar_graph_attribute("occupation")
# # model_adult.bar_graph_attribute("workclass")
# print(model_adult._train_data_set["capital-loss"])
#
# model_adult.normalise_attribute("capital-loss")
# model_adult.box_cox_trans_attribute("capital-loss", 5)
# model_adult.normalise_attribute("capital-loss")
# print(model_adult._train_data_set["capital-loss"])
# model_adult.histogram_and_q_q("capital-loss")
model_adult.move_target_to_train_y("salary")
model_adult.move_target_to_test_y('salary')
#print(model_adult._x_train)
# model_adult.random_forest()
model_adult.drop_attribute("fnlwgt")
model_adult.drop_attribute("workclass")
model_adult.drop_attribute("education-num")
model_adult.drop_attribute("marital-status")
model_adult.drop_attribute("occupation")
model_adult.one_hot_encode_attribute("relationship")
model_adult.one_hot_encode_attribute("education")
model_adult.drop_attribute("capital-gain")
model_adult.drop_attribute('capital-loss')
model_adult.drop_attribute("hours-per-week")
model_adult.one_hot_encode_attribute("race")
model_adult.one_hot_encode_attribute("sex")
model_adult.drop_attribute("native-country")
model_adult.drop_attribute('age')
print(model_adult._test_data_set)
#model_adult.random_forest()
model_adult.shuffle_data_set()
#print(model_adult._test_data_set)
model_adult.delete_unnecessary_one_hot_encoded_columns()
for i in range(len(model_adult._y_test.values)):
model_adult._y_test.values[i] = model_adult._y_test.values[i].strip('.')
my_random_forest_model = model_adult.random_forest()