Fire up graphlab create


In [2]:
import graphlab

Load some house sales data

Dataset is from house sales in King County, the region where the city of Seattle, WA is located.


In [4]:
sales = graphlab.SFrame('home_data.gl/')

In [5]:
sales


Out[5]:
id date price bedrooms bathrooms sqft_living sqft_lot floors waterfront
7129300520 2014-10-13 00:00:00+00:00 221900 3 1 1180 5650 1 0
6414100192 2014-12-09 00:00:00+00:00 538000 3 2.25 2570 7242 2 0
5631500400 2015-02-25 00:00:00+00:00 180000 2 1 770 10000 1 0
2487200875 2014-12-09 00:00:00+00:00 604000 4 3 1960 5000 1 0
1954400510 2015-02-18 00:00:00+00:00 510000 3 2 1680 8080 1 0
7237550310 2014-05-12 00:00:00+00:00 1225000 4 4.5 5420 101930 1 0
1321400060 2014-06-27 00:00:00+00:00 257500 3 2.25 1715 6819 2 0
2008000270 2015-01-15 00:00:00+00:00 291850 3 1.5 1060 9711 1 0
2414600126 2015-04-15 00:00:00+00:00 229500 3 1 1780 7470 1 0
3793500160 2015-03-12 00:00:00+00:00 323000 3 2.5 1890 6560 2 0
view condition grade sqft_above sqft_basement yr_built yr_renovated zipcode lat
0 3 7 1180 0 1955 0 98178 47.51123398
0 3 7 2170 400 1951 1991 98125 47.72102274
0 3 6 770 0 1933 0 98028 47.73792661
0 5 7 1050 910 1965 0 98136 47.52082
0 3 8 1680 0 1987 0 98074 47.61681228
0 3 11 3890 1530 2001 0 98053 47.65611835
0 3 7 1715 0 1995 0 98003 47.30972002
0 3 7 1060 0 1963 0 98198 47.40949984
0 3 7 1050 730 1960 0 98146 47.51229381
0 3 7 1890 0 2003 0 98038 47.36840673
long sqft_living15 sqft_lot15
-122.25677536 1340.0 5650.0
-122.3188624 1690.0 7639.0
-122.23319601 2720.0 8062.0
-122.39318505 1360.0 5000.0
-122.04490059 1800.0 7503.0
-122.00528655 4760.0 101930.0
-122.32704857 2238.0 6819.0
-122.31457273 1650.0 9711.0
-122.33659507 1780.0 8113.0
-122.0308176 2390.0 7570.0
[21613 rows x 21 columns]
Note: Only the head of the SFrame is printed.
You can use print_rows(num_rows=m, num_columns=n) to print more rows and columns.

Exploring the data for housing sales

The house price is correlated with the number of square feet of living space.


In [7]:
graphlab.canvas.set_target("ipynb")
sales.show(view="Scatter Plot", x="sqft_living", y="price")


Create a simple regression model of sqft_living to price

Split data into training and testing.

.8 = 80% of the data is for training and other 20% for test.

We use seed=0 so that everyone running this notebook gets the same results. In practice, you may set a random seed (or let GraphLab Create pick a random seed for you).


In [8]:
train_data,test_data = sales.random_split(.8,seed=0)

Build the regression model using only sqft_living as a feature


In [9]:
sqft_model = graphlab.linear_regression.create(train_data, target='price', features=['sqft_living'])


PROGRESS: Creating a validation set from 5 percent of training data. This may take a while.
          You can set ``validation_set=None`` to disable validation tracking.

PROGRESS: Linear regression:
PROGRESS: --------------------------------------------------------
PROGRESS: Number of examples          : 16531
PROGRESS: Number of features          : 1
PROGRESS: Number of unpacked features : 1
PROGRESS: Number of coefficients    : 2
PROGRESS: Starting Newton Method
PROGRESS: --------------------------------------------------------
PROGRESS: +-----------+----------+--------------+--------------------+----------------------+---------------+-----------------+
PROGRESS: | Iteration | Passes   | Elapsed Time | Training-max_error | Validation-max_error | Training-rmse | Validation-rmse |
PROGRESS: +-----------+----------+--------------+--------------------+----------------------+---------------+-----------------+
PROGRESS: | 1         | 2        | 1.020559     | 4378456.764083     | 4301917.860934       | 259886.756504 | 316634.252867   |
PROGRESS: +-----------+----------+--------------+--------------------+----------------------+---------------+-----------------+

Evaluate the simple model


In [10]:
print test_data['price'].mean()


543054.042563

evaluate() takes a test dataset and returns statistics about that set


In [11]:
print sqft_model.evaluate(test_data)


{'max_error': 4166280.6293561924, 'rmse': 255144.10764777666}

RMSE of about \$255,170!

Let's show what our predictions look like

Matplotlib is a Python plotting library that is also useful for plotting. You can install it with:

'pip install matplotlib'


In [13]:
import matplotlib.pyplot as plt
%matplotlib inline

In [17]:
# predict() outputs SArray column with predicted values
plt.plot(test_data['sqft_living'],test_data['price'],'+',
        test_data['sqft_living'],sqft_model.predict(test_data),'-')


Out[17]:
[<matplotlib.lines.Line2D at 0x10f3f05d0>,
 <matplotlib.lines.Line2D at 0x10f3f0810>]

Above: blue dots are original data, green line is the prediction from the simple regression.

Below: we can view the learned regression coefficients.


In [22]:
sqft_model.get('coefficients')


Out[22]:
name index value
(intercept) None -41432.4939646
sqft_living None 279.085122812
[2 rows x 3 columns]

Explore other features in the data

To build a more elaborate model, we will explore using more features.


In [23]:
my_features = ['bedrooms', 'bathrooms', 'sqft_living', 'sqft_lot', 'floors', 'zipcode']

In [24]:
sales[my_features].show()



In [25]:
sales.show(view='BoxWhisker Plot', x='zipcode', y='price')


Pull the bar at the bottom to view more of the data.

98039 is the most expensive zip code.

Build a regression model with more features


In [26]:
my_features_model = graphlab.linear_regression.create(train_data,target='price',features=my_features,validation_set=None)


PROGRESS: Linear regression:
PROGRESS: --------------------------------------------------------
PROGRESS: Number of examples          : 17384
PROGRESS: Number of features          : 6
PROGRESS: Number of unpacked features : 6
PROGRESS: Number of coefficients    : 115
PROGRESS: Starting Newton Method
PROGRESS: --------------------------------------------------------
PROGRESS: +-----------+----------+--------------+--------------------+---------------+
PROGRESS: | Iteration | Passes   | Elapsed Time | Training-max_error | Training-rmse |
PROGRESS: +-----------+----------+--------------+--------------------+---------------+
PROGRESS: | 1         | 2        | 0.054584     | 3763208.270524     | 181908.848367 |
PROGRESS: +-----------+----------+--------------+--------------------+---------------+

In [27]:
print my_features


['bedrooms', 'bathrooms', 'sqft_living', 'sqft_lot', 'floors', 'zipcode']

Comparing the results of the simple model with adding more features


In [28]:
print sqft_model.evaluate(test_data)
print my_features_model.evaluate(test_data)


{'max_error': 4166280.6293561924, 'rmse': 255144.10764777666}
{'max_error': 3486584.509381928, 'rmse': 179542.43331269105}

The RMSE goes down from \$255,170 to \$179,508 with more features.

Apply learned models to predict prices of 3 houses

The first house we will use is considered an "average" house in Seattle.


In [30]:
house1 = sales[sales['id']=='5309101200']

In [33]:
house1


Out[33]:
id date price bedrooms bathrooms sqft_living sqft_lot floors waterfront
5309101200 2014-06-05 00:00:00+00:00 620000 4 2.25 2400 5350 1.5 0
view condition grade sqft_above sqft_basement yr_built yr_renovated zipcode lat
0 4 7 1460 940 1929 0 98117 47.67632376
long sqft_living15 sqft_lot15
-122.37010126 1250.0 4880.0
[1 rows x 21 columns]


In [34]:
print house1['price']


[620000]

In [35]:
print sqft_model.predict(house1)


[628371.8007836473]

In [37]:
print my_features_model.predict(house1)


[721918.9333272816]

In this case, the model with more features provides a worse prediction than the simpler model with only 1 feature. However, on average, the model with more features is better.

Prediction for a second, fancier house

We will now examine the predictions for a fancier house.


In [38]:
house2 = sales[sales['id']=='1925069082']

In [39]:
house2


Out[39]:
id date price bedrooms bathrooms sqft_living sqft_lot floors waterfront
1925069082 2015-05-11 00:00:00+00:00 2200000 5 4.25 4640 22703 2 1
view condition grade sqft_above sqft_basement yr_built yr_renovated zipcode lat
4 5 8 2860 1780 1952 0 98052 47.63925783
long sqft_living15 sqft_lot15
-122.09722322 3140.0 14200.0
[? rows x 21 columns]
Note: Only the head of the SFrame is printed. This SFrame is lazily evaluated.
You can use len(sf) to force materialization.


In [40]:
print sqft_model.predict(house2)


[1253522.4758820129]

In [41]:
print my_features_model.predict(house2)


[1446472.4690774996]

In this case, the model with more features provides a better prediction. This behavior is expected here, because this house is more differentiated by features that go beyond its square feet of living space, especially the fact that it's a waterfront house.

Last house, super fancy

Our last house is a very large one owned by a famous Seattleite.


In [42]:
bill_gates = {'bedrooms':[8], 
              'bathrooms':[25], 
              'sqft_living':[50000], 
              'sqft_lot':[225000],
              'floors':[4], 
              'zipcode':['98039'], 
              'condition':[10], 
              'grade':[10],
              'waterfront':[1],
              'view':[4],
              'sqft_above':[37500],
              'sqft_basement':[12500],
              'yr_built':[1994],
              'yr_renovated':[2010],
              'lat':[47.627606],
              'long':[-122.242054],
              'sqft_living15':[5000],
              'sqft_lot15':[40000]}


In [43]:
print my_features_model.predict(graphlab.SFrame(bill_gates))


[13749825.525717655]

The model predicts a price of over $13M for this house! But we expect the house to cost much more. (There are very few samples in the dataset of houses that are this fancy, so we don't expect the model to capture a perfect prediction here.)

Improve the regression model

  1. Find zip code of area with highest sales and compute average price
  2. Filter houses by those within 2000 to 4000 square feet
  3. Compute error for model using my_features model and a newer model with advanced features.

In [50]:
highest_sales = sales[sales['zipcode']=='98039']

In [53]:
highest_sales_avg_price = highest_sales['price'].mean()

In [54]:
print highest_sales_avg_price


2160606.6

In [58]:
highest_sales_range = highest_sales[(highest_sales['sqft_living'] > 2000) & (highest_sales['sqft_living'] <= 4000)]

In [64]:
len(highest_sales_range) / float(50)


Out[64]:
0.48

In [65]:
advanced_features = [
    'bedrooms', 'bathrooms', 'sqft_living', 'sqft_lot', 'floors', 'zipcode',
    'condition',
    'grade',
    'waterfront',
    'view',
    'sqft_above',
    'sqft_basement',
    'yr_built',
    'yr_renovated',
    'lat', 'long',
    'sqft_living15',
    'sqft_lot15'
]

In [72]:
advanced_features_model = graphlab.linear_regression.create(train_data, target='price',features=advanced_features, validation_set=None)


PROGRESS: Linear regression:
PROGRESS: --------------------------------------------------------
PROGRESS: Number of examples          : 17384
PROGRESS: Number of features          : 18
PROGRESS: Number of unpacked features : 18
PROGRESS: Number of coefficients    : 127
PROGRESS: Starting Newton Method
PROGRESS: --------------------------------------------------------
PROGRESS: +-----------+----------+--------------+--------------------+---------------+
PROGRESS: | Iteration | Passes   | Elapsed Time | Training-max_error | Training-rmse |
PROGRESS: +-----------+----------+--------------+--------------------+---------------+
PROGRESS: | 1         | 2        | 0.032847     | 3469012.450663     | 154580.940735 |
PROGRESS: +-----------+----------+--------------+--------------------+---------------+

In [73]:
my_features_model.evaluate(test_data)


Out[73]:
{'max_error': 3486584.509381928, 'rmse': 179542.43331269105}

In [74]:
advanced_features_model.evaluate(test_data)


Out[74]:
{'max_error': 3556849.413848093, 'rmse': 156831.11680191013}

In [ ]: