def main():
loader = Loader(FILE_PATH, PREDICT_PATH)
loader.load_item_base()
item_base = ItemBase(loader, 5)
loader2 = Loader(FILE_PATH, PREDICT_PATH)
loader2.load_user_base()
item_base.user_data = loader2.data
try:
item_base.mean = pickle.load(open('temp/item_base/user_mean.p', 'rb'))
except Exception:
item_base.mean()
pickle.dump(item_base.mean, open('temp/item_base/user_mean.p', 'wb'))
try:
item_base.cosine = pickle.load(open('temp/item_base/ad_cosine.matrix', 'rb'))
except Exception:
item_base.similarities()
pickle.dump(item_base.cosine, open('temp/item_base/ad_cosine.matrix', 'wb'))
print '> Testing model'
k_values = [5, 10, 20, 30]
results = {}
results.setdefault('RMSE', [])
results.setdefault('Precision', [])
results['K'] = k_values
kf = cross_validation.KFold(len(item_base.loader.items), n_folds=5)
for k in k_values:
item_base.k = k
RMSE = 0.
precision = 0.
for train_index, test_index in kf:
item_base.train_idx = train_index
item_base.test_idx = test_index
item_base.knn()
RMSE += item_base.test_error(test_index)
precision += item_base.test_topN(test_index)
print str.format('RMSE [k={0}] = {1}', k, RMSE/5)
print str.format('PRECISION@10 [k={0}] = {1}', k, precision/5)
results['RMSE'].append(RMSE / 5)
results['Precision'].append(precision / 5)
print results
predictions = sorted(predictions, key=operator.itemgetter(1))
predictions.reverse()
top10[user] = predictions[0:10]
file = open("temp/slope_one/top10.txt", "wb")
for user,items in top10.items():
file.write(str.format("{0}\n", user))
for data in items:
file.write(str.format('\t"{0}"\n', data[0]))
if __name__ == '__main__':
loader = Loader(FILE_PATH, PREDICT_PATH)
loader.load_user_base()
slope_one = SlopeOne(loader)
loader2 = Loader(FILE_PATH, PREDICT_PATH)
loader2.load_item_base()
slope_one.item_data = loader2.data
try:
slope_one.avg = pickle.load(open('temp/slope_one/mean.p', 'rb'))
except Exception:
slope_one.mean()
pickle.dump(slope_one.avg, open('temp/slope_one/mean.p', 'wb'))
try:
print '> Loading dev matrix'
f = open('temp/slope_one/freqs.p', 'rb')
slope_one.freqs = pickle.load(f)
f.close()
f = open('temp/slope_one/diffs.p', 'rb')
slope_one.diffs = pickle.load(f)
