Kaggle

Table of contents:

• ML

ML

import pandas as pd
file_path = '../input/melbourne-housing-snapshot/melb_data.csv'
melbdata = pd.read_csv(file_path)
melbdata.describe()   # summary of data

1st ml model

import pandas as pd 
melbdata.columns      # listing all columns code

melbdatana = melbdata.dropna(axis=0)     # drop not available = missing values

y = melbdatana.Price     # dot notation = prediction target

melbdata_features = ['Rooms', 'Bathroom', 'Lattitude', 'Longtitude', 'Landsize']     # choosing features
X = melbdatana[melbdata_features]
X.describe()
X.head()

from sklearn.tree import DecisionTreeRegressor
#   define model, specify a number for random state to ensure same results each run
melbmodel = DecisionTreeRegressor(random_state=1)
melbmodel.fit(X,y)

print('Making predictions for the following 5 houses')
print(X.heaad())
print('The predictions are')
print(mebmodel.predict(X.head()))

model validation - measure quality of model

# On average, our predictions are off by about X.

from sklearn.model_selection import train_test_split
train_X, val_X, train_y, val_y = train_test_split(X,y, random_state = 0)
melbmodel = DecisionTreeRegressor()
melbmodel.fit(train_X, train_y)
val_predictions = melbmodel.predict(val_X)
print(mean_absolute_error(val_y, val_predictions))

underfitting and overfitting

# max_leaf_nodes - sensible way to control overfitting and underfitting
from sklearn.metrics import mean_absolute_error
from sklearn.tree import DecisionTreeRegressor
def get_mae(max_leaf_nodes, train_X, val_X, train_y, val_y):
    model = DecisionTreeRegressor(max_leaf_nodes=max_leaf_nodes, random_state=0)
    model.fit(train_X, train_Y)
    pred_val = model.predict(val_X)
    mae=mean_absolute_error(val_y, pred_val)
    return(mae)

# compare MAE with different values of max_leaf_nodes
for max_leaf_nodes in [5,50,500,5000]:
    my_mae = get_mae(max_leaf_nodes, train_X, val_X, train_y, val_y)
    print('Max leaf nodes: &d \t\t Mean absolute error: %d' %(max_leaf_nodes, my_mae))

Random Forests

# random forests uses many trees, and it makes a prediction by averaging the predictions of each componenet tree

from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_absolute_error

forestmodel = RandomForestRegressor
foresmodel.fit(train_X, train_y)
prediction = forestmodel.predict(val_X)
print(mean_absolute_error(val_y, prediction))