def from_hdf(fpath, dbg=None):
"""
Restores a Results object from HDF5 as created by the to_hdf method.
"""
if dbg is None:
dbg = Dbg()
import h5py
data = pd.read_hdf(fpath, 'data')
hfile = h5py.File(fpath)
md = hfile['metadata']
datetime_conversions = set(md.attrs.get('_datetime_attributes', set()))
metadata = {}
for k, v in list(md.attrs.items()):
if k == '_datetime_attributes':
continue
dbg(3, 'from_hdf: retrieving metadata {}'.format(k))
if k in datetime_conversions:
dbg(3, ' -> converting to Timestamp')
v = pd.Timestamp(v)
metadata[k] = v
return Results(data, metadata)
def __init__(self, df, metadata={}, dbg=None):
"""
Want to be able to create results from just a single dataframe.
"""
self.df = df
self.metadata = metadata
self.metadata['version'] = __version__
self.metadata['errors'] = {}
self.metadata['warnings'] = {}
self.dbg = dbg or Dbg()
def setUp(self):
"""
Load the needed datasets for all StatisticsTestCases and set the random
seed so that randomized algorithms show deterministic behaviour.
"""
np.random.seed(0)
self.data = Experiment('B', *generate_random_data(), dbg=Dbg(dbg_lvl=5))
# Create time column. TODO: Do this nicer
self.data.kpis['time_since_treatment'] = \
self.data.features['treatment_start_time']
# Make time part of index
self.data.kpis.set_index('time_since_treatment', append=True, inplace=True)
# Metadata as generated by generate_random_data() for later checks
self.testmetadata = {'primary_KPI': 'normal_shifted',
'source': 'simulated',
'experiment': 'random_data_generation'}
def __init__(self, baseline_variant, metrics_or_kpis, metadata={},
features='default', dbg=None):
# Call constructor of super class
super(Experiment, self).__init__(metrics_or_kpis, metadata, features)
# If no baseline variant is found
if ((baseline_variant not in self.kpis.index.levels[
self.primary_indices.index('variant')])
and (baseline_variant not in self.features.index.levels[
self.primary_indices.index('variant')])):
raise KeyError('baseline_variant ({}) not present in KPIs or features.'.format(
baseline_variant))
# Add baseline to metadata
self.metadata['baseline_variant'] = baseline_variant
self.dbg = dbg or Dbg()
def __init__(self, df, metadata={}, dbg=None):
"""
Want to be able to create results from just a single dataframe.
Args:
df (pandas.DataFrame): input dataframe
metadata (dict): input metadata
dbg:
"""
self.df = df
self.metadata = metadata
self.metadata['version'] = __version__
self.metadata['errors'] = {}
self.metadata['warnings'] = {}
self.dbg = dbg or Dbg()
def setUp(self):
"""
Load the needed datasets for all StatisticsTestCases and set the random
seed so that randomized algorithms show deterministic behaviour.
"""
np.random.seed(0)
self.data = Experiment('B',
*generate_random_data(),
dbg=Dbg(dbg_lvl=5))
# Create time column. TODO: Do this nicer
self.data.kpis['time_since_treatment'] = \
self.data.features['treatment_start_time']
# Make time part of index
self.data.kpis.set_index('time_since_treatment',
append=True,
inplace=True)