def __getitem__(self, index):
"Index by round number, list (or tuple), or slice"
if isinstance(index, slice):
if index.step is not None:
raise ValueError("slice step size must be 1")
r1, r2 = self.rounds_to_ints(index.start, index.stop)
rs = list(range(r1, r2 + 1))
ts = nx.tournament_all(as_str=False)
elif nx.isint(index):
rs = [index]
ts = nx.tournament_all(as_str=False)
elif isinstance(index, list):
rs, ts = zip(*index)
ts = [nx.tournament_int(i) for i in ts]
elif isinstance(index, tuple):
if len(index) != 2:
raise IndexError("tuple index must have length 2")
r, t = index
if not nx.isint(r):
raise IndexError("first element of tuple index must be int")
if not (nx.isint(t) or nx.isstring(t)):
msg = "second element of tuple index must be int or str"
raise IndexError(msg)
rs = [r]
ts = [nx.tournament_int(t)]
else:
raise IndexError("indexing method not supported")
self.gets(rs, ts)
ridx = self.df['round'].isin(rs)
tidx = self.df['tournament'].isin(ts)
idx = ridx & tidx
df = self.df[idx]
return df
def next_split(self):
if not nx.isint(self.p['seed']):
raise ValueError("`seed` must be an integer")
data = self.p['data']
if self.count == 0:
if self.p['train_only']:
data = data['train']
eras = data.unique_era()
n = int(eras.size / self.p['kfold'])
idx = self.p['seed'] % n
if idx == 0:
era0 = eras[idx:n]
eras = eras[n:]
else:
era0 = np.concatenate((eras[idx:n], eras[-idx:]))
eras = np.concatenate((eras[:idx], eras[n:-idx]))
eras = np.array_split(eras, self.p['kfold'] - 1)
eras.append(era0)
self.eras = eras
fit_eras = []
for i, e in enumerate(self.eras):
if i != self.count:
fit_eras.append(e)
fit_eras = np.concatenate(fit_eras)
dfit = data.era_isin(fit_eras)
pre_eras = self.eras[self.count]
dpre = data.era_isin(pre_eras)
return dfit, dpre
def tournament_str(tournament_int_or_str):
"Convert tournament int or str to str"
if nx.isstring(tournament_int_or_str):
if tournament_int_or_str not in nx.tournament_names(active_only=True):
raise ValueError('tournament name is unknown')
return tournament_int_or_str
elif nx.isint(tournament_int_or_str):
return tournament_int2str(tournament_int_or_str)
raise ValueError('input must be a str or int')
def tournament_str(tournament_int_or_str):
"""Convert tournament int or str to str"""
if nx.isstring(tournament_int_or_str):
if tournament_int_or_str not in nx.tournament_names(active_only=False):
raise ValueError('tournament name is not recognized')
return tournament_int_or_str
elif nx.isint(tournament_int_or_str):
return tournament_int2str(tournament_int_or_str)
raise ValueError('input must be a str or int')
def tournament_int(tournament_int_or_str):
"Convert tournament int or str to int"
if nx.isstring(tournament_int_or_str):
return tournament_str2int(tournament_int_or_str)
elif nx.isint(tournament_int_or_str):
if tournament_int_or_str not in (1, 2, 3, 4, 5):
raise ValueError('tournament int must be between 1 and 5')
return tournament_int_or_str
raise ValueError('input must be a str or int')
def tournament_str(tournament_int_or_str):
"Convert tournament int or str to str"
if nx.isstring(tournament_int_or_str):
if tournament_int_or_str not in TOURNAMENT_NAMES:
raise ValueError('tournament name is unknown')
return tournament_int_or_str
elif nx.isint(tournament_int_or_str):
return tournament_int2str(tournament_int_or_str)
raise ValueError('input must be a str or int')
def tournament_int(tournament_int_or_str):
"Convert tournament int or str to int"
if nx.isstring(tournament_int_or_str):
return tournament_str2int(tournament_int_or_str)
elif nx.isint(tournament_int_or_str):
numbers = nx.tournament_numbers(active_only=True)
if tournament_int_or_str not in numbers:
raise ValueError("`tournament_int_or_str` not recognized")
return tournament_int_or_str
raise ValueError('input must be a str or int')
def test_isint():
"test isint"
ok_(nx.isint(1))
ok_(nx.isint(-1))
ok_(not nx.isint(1.1))
ok_(not nx.isint('a'))
ok_(not nx.isint(True))
ok_(not nx.isint(False))
ok_(not nx.isint(None))
def whatif(self, users, s, c, round1, round2):
"""
Profit if `users` had staked `s` and `c` in every tournament.
Earnings are left in NMR instead of splitting the NMR earnings into
NMR and USD.
"""
if nx.isint(round1):
if round1 < 113:
raise ValueError("`round1` must start at at least 113")
lb = self.lb[round1:round2]
df = whatif(lb, users, s, c)
return df
def pairs_with_tournament(self, tournament, as_str=True):
"""List of pairs; `tournament` can be int, str, or list"""
if isinstance(tournament, list):
tournaments = [nx.tournament_int(t) for t in tournament]
elif nx.isstring(tournament):
tournaments = [nx.tournament_int(tournament)]
elif nx.isint(tournament):
tournaments = [tournament]
else:
raise ValueError('`tournament` must be int, str or list')
prs = self.pairs(as_str=False)
pairs = []
for pr in prs:
if pr[1] in tournaments:
if as_str:
pr = (pr[0], nx.tournament_str(pr[1]))
pairs.append(pr)
return pairs
def __getitem__(self2, index):
if isinstance(index, str):
if index in ('bernie', 'elizabeth', 'jordan', 'ken', 'charles'):
return self2.df[index].values
else:
raise IndexError('string index not recognized')
elif nx.isint(index):
if index < 1 or index > 5:
raise IndexError('tournament number must be between 1 and 5')
return self2.df[nx.tournament_str(index)].values
elif isinstance(index, slice):
if (index.start is None and index.stop is None
and index.step is None):
# slicing below means a view is returned instead of a copy
return self2.df.iloc[:, -5:].values
else:
raise IndexError('Start, stop, and step of slice must be None')
else:
raise IndexError('indexing type not recognized')
def __getitem__(self, index):
"Prediction indexing is by model pair(s)"
if isinstance(index, tuple):
if len(index) != 2:
raise IndexError("When indexing by tuple must be length 2")
if isinstance(index[0], slice):
if not is_none_slice(index[0]):
raise IndexError("Slces must be slice(None, None, None,)")
pairs1 = self.pairs(as_str=False)
elif nx.isstring(index[0]):
pairs1 = self.pairs_with_name(index[0], as_str=False)
else:
raise IndexError("indexing method not recognized")
if isinstance(index[1], slice):
if not is_none_slice(index[1]):
raise IndexError("Slces must be slice(None, None, None,)")
pairs2 = self.pairs(as_str=False)
elif nx.isint(index[1]):
pairs2 = self.pairs_with_tournament(index[1], as_str=False)
elif nx.isstring(index[1]):
pairs2 = self.pairs_with_tournament(index[1], as_str=False)
else:
raise IndexError("indexing method not recognized")
pairs = []
for pair in pairs1:
if pair in pairs2:
pairs.append(pair)
p = Prediction(pd.DataFrame(data=self.df[pairs]))
elif nx.isstring(index):
pairs = self.pairs_with_name(index, as_str=False)
p = Prediction(self.df[pairs])
else:
# assume an iterable of tuple pairs
idx = []
for ix in index:
if len(ix) != 2:
msg = "Expecting list of tuple pairs with length 2"
raise IndexError(msg)
idx.append((ix[0], nx.tournament_int(ix[1])))
p = Prediction(self.df[idx])
return p
def __getitem__(self2, index):
n = nx.tournament_count(active_only=False)
if isinstance(index, str):
if index in nx.tournament_all(as_str=True, active_only=True):
return self2.df[index].values
else:
raise IndexError('string index not recognized')
elif nx.isint(index):
if index < 1 or index > n:
txt = 'tournament number must be between 1 and {}'
raise IndexError(txt.format(n))
return self2.df[nx.tournament_str(index)].values
elif isinstance(index, slice):
if (index.start is None and index.stop is None
and index.step is None):
# slicing below means a view is returned instead of a copy
return self2.df.iloc[:, -n:].values
else:
raise IndexError('Start, stop, and step of slice must be None')
else:
raise IndexError('indexing type not recognized')
def __getitem__(self, index):
"Index by round number, list (or tuple), or slice"
if isinstance(index, slice):
if index.step is not None:
raise ValueError("slice step size must be 1")
r1, r2 = self.rounds_to_ints(index.start, index.stop)
self.get_range(r1, r2)
idx1 = self.df['round'] >= r1
idx2 = self.df['round'] <= r2
idx = idx1 & idx2
df = self.df[idx]
elif nx.isint(index):
self.get_round(index)
df = self.df[self.df['round'] == index]
elif isinstance(index, list) or isinstance(index, tuple):
self.get_list(index)
idx = self.df['round'].isin(index)
df = self.df[idx]
else:
raise IndexError("indexing method not supported")
return df
def run(model, splitter, tournament=None, verbosity=2):
"""
Run a model/tournament pair (or pairs) through a data splitter.
Parameters
----------
model : nx.Model, list, tuple
Prediction model to run through the splitter. Can be a list or tuple
of prediction models. Model names must be unique.
splitter : nx.Splitter
An iterator of fit/predict data pairs.
tournament : {None, int, str, list, tuple}, optional
The tournament(s) to run the model through. By default (None) the
model is run through all active tournaments. If a list or tuple of
tournaments is given then it must must not contain duplicate
tournaments.
verbosity : int, optional
An integer that determines verbosity. Zero is silent.
Returns
-------
p : nx.Prediction
A prediction object containing the predictions of the specified
model/tournament pairs.
"""
# make list of models
if isinstance(model, nx.Model):
models = [model]
elif isinstance(model, list) or isinstance(model, tuple):
models = model
else:
raise ValueError('`model` must be a model, list, or tuple of models')
names = [m.name for m in models]
if len(names) != len(set(names)):
raise ValueError('`model` cannot contain duplicate names')
# make list of tournaments
if tournament is None:
tournaments = nx.tournament_all()
elif nx.isint(tournament) or nx.isstring(tournament):
tournaments = [tournament]
elif isinstance(tournament, list) or isinstance(tournament, tuple):
tournaments = tournament
else:
msg = '`tournament` must be an integer, string, list, tuple, or None.'
raise ValueError(msg)
tournaments = [nx.tournament_str(t) for t in tournaments]
if len(tournaments) != len(set(tournaments)):
raise ValueError('`tournament` cannot contain duplicates')
# loop over all model/tournament pairs
p = nx.Prediction()
for m in models:
for t in tournaments:
p += run_one(m, splitter, t, verbosity=verbosity)
splitter.reset()
splitter.reset()
return p
def __getitem__(self, index):
"""Data indexing"""
typidx = type(index)
if isinstance(index, str):
if index.startswith('era'):
if len(index) < 4:
raise IndexError('length of era string index too short')
return self.era_isin([index])
else:
if index in ('train', 'validation', 'test', 'live'):
return self.region_isin([index])
elif index == 'tournament':
return self.region_isin(TOURNAMENT_REGIONS)
else:
raise IndexError('string index not recognized')
elif isinstance(index, slice):
# step check
if index.step is not None:
if not nx.isint(index.step):
msg = "slice step size must be None or psotive integer"
raise IndexError(msg)
if index.step < 1:
raise IndexError('slice step must be greater than 0')
step = index.step
else:
step = 1
ueras = self.unique_era().tolist()
# start
era1 = index.start
idx1 = None
if era1 is None:
idx1 = 0
elif not nx.isstring(era1) or not era1.startswith('era'):
raise IndexError("slice elements must be strings like 'era23'")
if idx1 is None:
idx1 = ueras.index(era1)
# end
era2 = index.stop
idx2 = None
if era2 is None:
idx2 = len(ueras) - 1
elif not nx.isstring(era2) or not era2.startswith('era'):
raise IndexError("slice elements must be strings like 'era23'")
if idx2 is None:
idx2 = ueras.index(era2)
if idx1 > idx2:
raise IndexError("slice cannot go from large to small era")
# find eras in slice
eras = []
for ix in range(idx1, idx2 + 1, step):
eras.append(ueras[ix])
data = self.era_isin(eras)
return data
elif typidx is pd.Series or typidx is np.ndarray:
return Data(self.df[index])
else:
raise IndexError('indexing type not recognized')
