def test_save(self):
t = XFrame({'id': [30, 20, 10], 'val': ['a', 'b', 'c']})
path = '{}/tmp/frame'.format(hdfs_prefix)
t.save(path, format='binary')
with fileio.open_file(os.path.join(path, '_metadata')) as f:
metadata = pickle.load(f)
self.assertListEqual([['id', 'val'], [int, str]], metadata)
# TODO find some way to check the data
fileio.delete(path)
def test_construct_auto_str_psv(self):
path = '{}/user/xpatterns/files/test-frame.psv'.format(hdfs_prefix)
res = XFrame(path)
self.assertEqualLen(3, res)
self.assertListEqual(['id', 'val'], res.column_names())
self.assertListEqual([int, str], res.column_types())
self.assertDictEqual({'id': 1, 'val': 'a'}, res[0])
self.assertDictEqual({'id': 2, 'val': 'b'}, res[1])
self.assertDictEqual({'id': 3, 'val': 'c'}, res[2])
def test_construct_str_xframe(self):
# construct and XFrame given a saved xframe
path = '{}/user/xpatterns/files/test-frame'.format(hdfs_prefix)
res = XFrame(path, format='xframe')
res = res.sort('id')
self.assertEqualLen(3, res)
self.assertListEqual(['id', 'val'], res.column_names())
self.assertListEqual([int, str], res.column_types())
self.assertDictEqual({'id': 1, 'val': 'a'}, res[0])
self.assertDictEqual({'id': 2, 'val': 'b'}, res[1])
self.assertDictEqual({'id': 3, 'val': 'c'}, res[2])
def test_construct_str_csv(self):
# construct and XFrame given a text file
# interpret as csv
path = '{}/user/xpatterns/files/test-frame.txt'.format(hdfs_prefix)
res = XFrame(path, format='csv')
self.assertEqualLen(3, res)
self.assertListEqual(['id', 'val'], res.column_names())
self.assertListEqual([int, str], res.column_types())
self.assertDictEqual({'id': 1, 'val': 'a'}, res[0])
self.assertDictEqual({'id': 2, 'val': 'b'}, res[1])
self.assertDictEqual({'id': 3, 'val': 'c'}, res[2])
def test_construct_auto_str_xframe(self):
# construct an XFrame given a file with unrecognized file extension
path = '{}/user/xpatterns/files/test-frame'.format(hdfs_prefix)
res = XFrame(path)
res = res.sort('id')
self.assertEqualLen(3, res)
self.assertListEqual(['id', 'val'], res.column_names())
self.assertListEqual([int, str], res.column_types())
self.assertDictEqual({'id': 1, 'val': 'a'}, res[0])
self.assertDictEqual({'id': 2, 'val': 'b'}, res[1])
self.assertDictEqual({'id': 3, 'val': 'c'}, res[2])
def test_save(self):
t = XFrame({'id': [30, 20, 10], 'val': ['a', 'b', 'c']})
path = '{}/tmp/frame-csv'.format(hdfs_prefix)
t.save(path, format='csv')
with fileio.open_file(path + '.csv') as f:
heading = f.readline().rstrip()
self.assertEqual('id,val', heading)
self.assertEqual('30,a', f.readline().rstrip())
self.assertEqual('20,b', f.readline().rstrip())
self.assertEqual('10,c', f.readline().rstrip())
fileio.delete(path + '.csv')
def test_construct_auto_dataframe(self):
path = '{}/user/xpatterns/files/test-frame-auto.csv'.format(hdfs_prefix)
res = XFrame(path)
self.assertEqualLen(3, res)
self.assertListEqual(['val_int', 'val_int_signed', 'val_float', 'val_float_signed',
'val_str', 'val_list', 'val_dict'], res.column_names())
self.assertListEqual([int, int, float, float, str, list, dict], res.column_types())
self.assertDictEqual({'val_int': 1, 'val_int_signed': -1, 'val_float': 1.0, 'val_float_signed': -1.0,
'val_str': 'a', 'val_list': ['a'], 'val_dict': {1: 'a'}}, res[0])
self.assertDictEqual({'val_int': 2, 'val_int_signed': -2, 'val_float': 2.0, 'val_float_signed': -2.0,
'val_str': 'b', 'val_list': ['b'], 'val_dict': {2: 'b'}}, res[1])
self.assertDictEqual({'val_int': 3, 'val_int_signed': -3, 'val_float': 3.0, 'val_float_signed': -3.0,
'val_str': 'c', 'val_list': ['c'], 'val_dict': {3: 'c'}}, res[2])
def load(cls, path):
"""
Load a model that was saved previously.
Parameters
----------
path : str
The path where the model files are stored.
This is the same path that was passed to ``save``.
There are three files/directories based on this path, with
extensions '.model', '.ratings', and '.metadata'.
Returns
-------
out : MatrixFactorizationModel
A model that can be used to predict ratings.
"""
sc = CommonSparkContext.Instance().sc()
model_path, ratings_path, metadata_path = cls._file_paths(path)
# load model
model = recommendation.MatrixFactorizationModel.load(sc, model_path)
# load ratings
ratings = XFrame.load(ratings_path)
# load metadata
with open(metadata_path) as f:
user_col, item_col, rating_col = pickle.load(f)
return cls(model, ratings, user_col, item_col, rating_col)
def test_read_text(self):
path = '{}/user/xpatterns/files/test-frame-text.txt'.format(hdfs_prefix)
res = XFrame.read_text(path)
self.assertEqualLen(3, res)
self.assertListEqual(['text', ], res.column_names())
self.assertListEqual([str], res.column_types())
self.assertDictEqual({'text': 'This is a test'}, res[0])
self.assertDictEqual({'text': 'of read_text.'}, res[1])
self.assertDictEqual({'text': 'Here is another sentence.'}, res[2])
def test_read_parquet_str(self):
t = XFrame({'id': [1, 2, 3], 'val': ['a', 'b', 'c']})
path = '{}/tmp/frame-parquet'.format(hdfs_prefix)
t.save(path, format='parquet')
res = XFrame('{}/tmp/frame-parquet.parquet'.format(hdfs_prefix))
# results may not come back in the same order
res = res.sort('id')
self.assertEqualLen(3, res)
self.assertListEqual(['id', 'val'], res.column_names())
self.assertListEqual([int, str], res.column_types())
self.assertDictEqual({'id': 1, 'val': 'a'}, res[0])
self.assertDictEqual({'id': 2, 'val': 'b'}, res[1])
self.assertDictEqual({'id': 3, 'val': 'c'}, res[2])
fileio.delete(path)
def __init__(self, features, labels, standardize=False):
self.standardize = standardize
self.means = None
self.stdevs = None
if standardize:
self.features = self._standardize(features)
else:
self.features = features
self.labels = labels
self.feature_cols = features.column_names()
labeled_feature_vector = XFrame(features)
label_col = 'label' # TODO what if there is a feature with this name ?
feature_cols = self.feature_cols # need local reference
labeled_feature_vector[label_col] = labels
def build_labeled_features(row):
label = row[label_col]
features =[row[col] for col in feature_cols]
return LabeledPoint(label, features)
self.labeled_feature_vector = labeled_feature_vector.apply(build_labeled_features)
def predict_all(self, user):
"""
Predict ratings for all items.
Parameters
----------
user : int
The user to make predictions for.
Returns
-------
out : XFrame
Each row of the frame consists of a user id, an item id, and a predicted rating.
"""
# build rdd to pass to predictAll
user_item = XFrame()
user_item[self.item_col] = self.items
user_item[self.user_col] = user
user_item.swap_columns(self.item_col, self.user_col)
rdd = user_item.to_rdd()
res = self.model.predictAll(rdd)
res = res.map(lambda rating: (rating.user, rating.product, rating.rating))
col_names = [self.user_col, self.item_col, self.rating_col]
user_type = self.users.dtype()
item_type = self.items.dtype()
col_types = [user_type, item_type, float]
return XFrame.from_rdd(res, column_names=col_names, column_types=col_types)
def _base_evaluate(self, data, labels):
"""
Evaluate the performance of the classifier.
Use the data to make predictions, then test the effectiveness of
the predictions against the labels.
The data must be a collection of items (XArray of SenseVector).
Returns
-------
out : A list of:
- overall correct prediction proportion
- true positive proportion
- true negative proportion
- false positive proportion
- false negative proportion
"""
results = XFrame()
predictions = self._base_predict(data)
results['predicted'] = predictions
results['actual'] = labels
# print results
def evaluate(row):
prediction = row['predicted']
actual = row['actual']
return {'correct': 1 if prediction == actual else 0,
'true_pos': 1 if prediction == 1 and actual == 1 else 0,
'true_neg': 1 if prediction == 0 and actual == 0 else 0,
'false_pos': 1 if prediction == 1 and actual == 0 else 0,
'false_neg': 1 if prediction == 0 and actual == 1 else 0,
'positive': 1 if actual == 1 else 0,
'negative': 1 if actual == 0 else 0
}
score = results.apply(evaluate)
def sum_item(item):
return score.apply(lambda x: x[item]).sum()
all_scores = float(len(labels))
correct = float(sum_item('correct'))
tp = float(sum_item('true_pos'))
tn = float(sum_item('true_neg'))
fp = float(sum_item('false_pos'))
fn = float(sum_item('false_neg'))
pos = float(sum_item('positive'))
neg = float(sum_item('negative'))
# precision = true pos / (true pos + false pos)
# recall = true pos / (true pos + false neg)
# true pos rate = true pos / positive
# false pos rate = false pos / negative
result = {}
result['correct'] = correct
result['true_pos'] = tp
result['true_neg'] = tn
result['false_pos'] = fp
result['false_neg'] = fn
result['all'] = all_scores
result['accuracy'] = correct / all_scores if all_scores > 0 else float('nan')
result['precision'] = tp / (tp + fp) if (tp + fp) > 0 else float('nan')
result['recall'] = tp / (tp + fn) if (tp + fn) > 0 else float('nan')
result['tpr'] = tp / pos if pos > 0 else float('nan')
result['fpr'] = fp / neg if neg > 0 else float('nan')
return result
from xframes import XFrame
xf = XFrame({'id': [1, 2, 3], 'val': ['a', 'b', 'c']})
print xf
def test_save(self):
t = XFrame({'id': [30, 20, 10], 'val': ['a', 'b', 'c']})
path = '{}/tmp/frame-parquet'.format(hdfs_prefix)
t.save(path, format='parquet')
# TODO verify
fileio.delete(path + '.parquet')