def test_PriceEstimator_create_prices_lstsq_soln_1():
"Test creation of price records with real data."
from clair.coredata import DataStore
from clair.prices import PriceEstimator
print "start"
data = DataStore()
data.read_data(relative("../../example-data"))
#Use all data as test data
# listings = data.listings
product_ids = [p.id for p in data.products
if not p.id.startswith("xxx-unknown")]
# #Take a small amount of test data.
listings = data.listings.ix[0:200]
# product_ids = [u'nikon-d70', u'nikon-d90', u'nikon-sb-24', u'nikon-sb-26',
# u'nikon-18-70-f/3.5-4.5--1', u'nikon-18-105-f/3.5-5.6--1',
# u'nikon-28-85-f/3.5-4.5--1']
print listings
# print listings.to_string(columns=["products", "price"])
estimator = PriceEstimator()
#Create matrix and vectors for linear least square
matrix, listing_prices, listing_ids, product_ids = \
estimator.compute_product_occurrence_matrix(listings, product_ids)
# print
# print "matrix:\n", matrix
# print "matrix rank:", np.linalg.matrix_rank(matrix)
# print "number products:", len(product_ids)
# print "listing_prices:\n", listing_prices
# print "listing_ids:\n", listing_ids
# print "product_ids:\n", product_ids
#Compute average product prices
product_prices, good_rows, good_cols, problem_products = \
estimator.solve_prices_lstsq(matrix, listing_prices,
listing_ids, product_ids)
#Create price records
prices = estimator.create_prices_lstsq_soln(matrix,
listing_prices, listing_ids,
product_prices, product_ids,
good_rows, good_cols, listings)
# print prices.to_string()
#TODO: assertions
print "finshed"
def test_PriceEstimator_solve_prices_lstsq_1():
"Test linear least square algorithm with real data."
from clair.coredata import DataStore
from clair.prices import PriceEstimator
print "start"
data = DataStore()
data.read_data(relative("../../example-data"))
#Take a small amount of test data.
listings = data.listings.ix[0:50]
# listings = data.listings
# product_ids = [p.id for p in data.products]
product_ids = [u'nikon-d70', u'nikon-d90', u'nikon-sb-24', u'nikon-sb-26',
u'nikon-18-70-f/3.5-4.5--1', u'nikon-18-105-f/3.5-5.6--1',
u'nikon-28-85-f/3.5-4.5--1']
print listings
print listings.to_string(columns=["products", "price"])
estimator = PriceEstimator()
#Create matrix and vectors for linear least square
matrix, listing_prices, listing_ids, product_ids = \
estimator.compute_product_occurrence_matrix(listings, product_ids)
print
print "matrix:\n", matrix
print "matrix rank:", np.linalg.matrix_rank(matrix)
print "number products:", len(product_ids)
print "listing_prices:\n", listing_prices
print "listing_ids:\n", listing_ids
print "product_ids:\n", product_ids
product_prices, good_rows, good_cols, problem_products = \
estimator.solve_prices_lstsq(matrix, listing_prices,
listing_ids, product_ids)
print "product_prices:\n", product_prices * 0.7
#TODO: assertions
print "finshed"
def test_PriceEstimator_create_prices_lstsq_soln_2():
"""
Test creation of price records from solution of linear
least square problem, with artificial data.
"""
from clair.prices import PriceEstimator
def print_vals():
print "matrix:\n", matrix
print "matrix rank:", np.linalg.matrix_rank(matrix)
print "number products:", len(product_ids)
print "listing_prices:\n", listing_prices
print "listing_ids:\n", listing_ids
print "product_ids:\n", product_ids
print "product_prices:\n", product_prices
print "real_prices:\n", real_prices
print "start"
estimator = PriceEstimator()
#Listing IDs, unimportant in this test.
listing_ids = array(["l1", "l2", "l3", "l4", "l5",
"l6", "l7", "l8", "l9", "l10"])
#Product IDs, and "real" prices for checking errors
product_ids = array(["a", "b", "c", "d", "e"])
real_prices = array([500, 200, 100, 50., 5.])
print "Matrix has full rank, no noise ---------------------------------"
#Matrix that represents the listings, each row is a listing
matrix = array([[ 1., 0., 0., 0., 0.,],
[ 1., 0., 0., 0., 0.,],
[ 0., 1., 0., 0., 0.,],
[ 0., 1., 0., 0., 0.,],
[ 1., 1., 0., 0., 0.,],
[ 1., 0., 1., 0., 0.,],
[ 0., 0., 1., 1., 0.,],
[ 0., 0., 1., 0., 1.,],
[ 0., 0., 0., 1., 1.,],
[ 1., 1., 1., 1., 1.,],
])
#compute listing prices from the real prices
listing_prices = dot(matrix, real_prices)
#Compute the product prices
product_prices, good_rows, good_cols, problem_products = \
estimator.solve_prices_lstsq(matrix, listing_prices,
listing_ids, product_ids)
print_vals()
prices = estimator.create_prices_lstsq_soln(matrix,
listing_prices, listing_ids,
product_prices, product_ids,
good_cols, good_rows)
print "prices:\n", prices.to_string()
true_prices = prices["price"] / prices["condition"]
prices_a = true_prices[prices["product"] == "a"]
prices_b = true_prices[prices["product"] == "b"]
prices_c = true_prices[prices["product"] == "c"]
prices_d = true_prices[prices["product"] == "d"]
prices_e = true_prices[prices["product"] == "e"]
np.testing.assert_allclose(prices_a, 500)
np.testing.assert_allclose(prices_b, 200)
np.testing.assert_allclose(prices_c, 100)
np.testing.assert_allclose(prices_d, 50)
np.testing.assert_allclose(prices_e, 5)
print "Matrix is 1*1 (but has full rank, no noise) ------------------------"
#Listing IDs, unimportant in this test.
listing_ids = array(["l1"])
#Product IDs, and "real" prices for checking errors
product_ids = array(["a"])
real_prices = array([500])
#Matrix that represents the listings, each row is a listing
matrix = array([[0.7]])
#compute listing prices from the real prices
listing_prices = dot(matrix, real_prices)
#Compute the product prices
product_prices, good_rows, good_cols, problem_products = \
estimator.solve_prices_lstsq(matrix, listing_prices,
listing_ids, product_ids)
print_vals()
prices = estimator.create_prices_lstsq_soln(matrix,
listing_prices, listing_ids,
product_prices, product_ids,
good_cols, good_rows)
print "prices:\n", prices.to_string()
true_prices = prices["price"] / prices["condition"]
prices_a = true_prices[prices["product"] == "a"]
np.testing.assert_allclose(prices_a, 500)
def test_PriceEstimator_solve_prices_lstsq_2():
"Test linear least square algorithm with artificial data."
from clair.prices import PriceEstimator
def print_vals():
print "matrix:\n", matrix
print "matrix rank:", np.linalg.matrix_rank(matrix)
print "number products:", len(product_ids)
print "listing_prices:\n", listing_prices
print "listing_ids:\n", listing_ids
print "product_ids:\n", product_ids
print "product_prices:\n", product_prices
print "real_prices:\n", real_prices
print "good_rows:\n", good_rows
print "good_cols:\n", good_cols
print "problem_products:\n", problem_products
print "start"
estimator = PriceEstimator()
#Listing IDs, unimportant in this test.
listing_ids = array(["l1", "l2", "l3", "l4", "l5",
"l6", "l7", "l8", "l9", "l10"])
#Product IDs, and "real" prices for checking errors
product_ids = array(["a", "b", "c", "d", "e"])
real_prices = array([500, 200, 100, 50., 5.])
print "Matrix has full rank, no noise ---------------------------------"
#Matrix that represents the listings, each row is a listing
matrix = array([[ 1., 0., 0., 0., 0.,],
[ 1., 0., 0., 0., 0.,],
[ 0., 1., 0., 0., 0.,],
[ 0., 1., 0., 0., 0.,],
[ 1., 1., 0., 0., 0.,],
[ 1., 0., 1., 0., 0.,],
[ 0., 0., 1., 1., 0.,],
[ 0., 0., 1., 0., 1.,],
[ 0., 0., 0., 1., 1.,],
[ 1., 1., 1., 1., 1.,],
])
#compute listing prices from the real prices
listing_prices = dot(matrix, real_prices)
#Compute the product prices
product_prices, good_rows, good_cols, problem_products = \
estimator.solve_prices_lstsq(matrix, listing_prices,
listing_ids, product_ids)
print_vals()
np.testing.assert_allclose(product_prices, real_prices)
print "\nMatrix has full rank, with noise --------------------------------"
#compute listing prices with noise
listing_prices = dot(matrix, real_prices)
listing_prices += np.random.normal(0, 0.1, (10,)) * listing_prices
#Compute the product prices
product_prices, good_rows, good_cols, problem_products = \
estimator.solve_prices_lstsq(matrix, listing_prices,
listing_ids, product_ids)
print_vals()
err_norm = norm(product_prices - real_prices)
print "Error norm:", err_norm
res_good = np.asarray(abs(product_prices - real_prices)
< real_prices * 0.2, dtype=int)
print "Number of results exact to 20%:", sum(res_good)
#This test might occasionally fail because current noise is too large.
assert sum(res_good) >= 3
print "\nMatrix has insufficient rank, no noise ---------------------------------"
#Matrix that represents the listings, each row is a listing
matrix = array([[ 1., 0., 0., 0., 0.,],
[ 1., 0., 0., 0., 0.,],
[ 0., 1., 0., 0., 0.,],
[ 0., 1., 0., 0., 0.,],
[ 1., 1., 0., 0., 0.,],
[ 1., 0., 1., 0., 0.,],
[ 0., 0., 0., 1., 1.,],
[ 0., 0., 0., 1., 1.,],
[ 0., 0., 0., 1., 1.,],
[ 1., 1., 1., 0., 0.,],
])
#compute listing prices from the real prices
listing_prices = dot(matrix, real_prices)
#Compute the product prices
product_prices, good_rows, good_cols, problem_products = \
estimator.solve_prices_lstsq(matrix, listing_prices,
listing_ids, product_ids)
print_vals()
np.testing.assert_allclose(product_prices[0:3], real_prices[0:3])
print "\nMatrix has insufficient rank, no noise ---------------------------------"
#Pathological case for the current algorithm
matrix = array([[ 1., 0., 0., 1., 1.,],
[ 0., 1., 0., 1., 1.,],
[ 0., 0., 1., 1., 1.,],
[ 0., 0., 0., 1., 1.,],
[ 0., 0., 0., 1., 1.,],
[ 0., 0., 0., 1., 1.,],
[ 0., 0., 0., 1., 1.,],
[ 0., 0., 0., 1., 1.,],
[ 0., 0., 0., 1., 1.,],
[ 0., 0., 0., 1., 1.,], ])
#compute listing prices from the real prices
listing_prices = dot(matrix, real_prices)
#Compute the product prices
product_prices, good_rows, good_cols, problem_products = \
estimator.solve_prices_lstsq(matrix, listing_prices,
listing_ids, product_ids)
print_vals()
np.testing.assert_allclose(product_prices[0:3], real_prices[0:3])
print "Matrix is 1*2 ------------------------"
#Listing IDs, unimportant in this test.
listing_ids = array(["l1"])
#Product IDs, and "real" prices for checking errors
product_ids = array(["a", "b"])
real_prices = array([500, 200.])
#Matrix that represents the listings, each row is a listing
matrix = array([[0.7, 0]])
#compute listing prices from the real prices
listing_prices = dot(matrix, real_prices)
#Compute the product prices
product_prices, good_rows, good_cols, problem_products = \
estimator.solve_prices_lstsq(matrix, listing_prices,
listing_ids, product_ids)
print_vals()
np.testing.assert_allclose(product_prices[good_cols],
real_prices[good_cols])
print "Matrix is 1*1 (but has full rank, no noise) ------------------------"
#Listing IDs, unimportant in this test.
listing_ids = array(["l1"])
#Product IDs, and "real" prices for checking errors
product_ids = array(["a"])
real_prices = array([500])
#Matrix that represents the listings, each row is a listing
matrix = array([[0.7]])
#compute listing prices from the real prices
listing_prices = dot(matrix, real_prices)
#Compute the product prices
product_prices, good_rows, good_cols, problem_products = \
estimator.solve_prices_lstsq(matrix, listing_prices,
listing_ids, product_ids)
print_vals()
np.testing.assert_allclose(product_prices[good_cols],
real_prices[good_cols])
print "finshed"