simonvpe
commented
9 years ago I also have the same problem. The tutorial example with AAPL data from earlier than 2010 seems to generate this, and other stocks too. A also tried manually downloading with pandas DataReader and get the same problem. There are no NAN or None in the data, that I verified with DataFrame.count. Please help, I`ve invested lots of lots of time in my algo and now I need to start testing it on more data for longer times.
BR Simon
---------------------------------------------------------------------------
LinAlgError Traceback (most recent call last)
<ipython-input-59-d9c5d6224029> in <module>()
2 #data2 = load_bars_from_yahoo(stocks=[sym], start=date_range[0], end=date_range[1])
3 algo = Trade()
----> 4 perf = algo.run(data, benchmark_return_source=False)
/usr/local/lib/python2.7/dist-packages/zipline-0.7.0-py2.7.egg/zipline/algorithm.pyc in run(self, source, overwrite_sim_params, benchmark_return_source)
474 # perf dictionary
475 perfs = []
--> 476 for perf in self.gen:
477 perfs.append(perf)
478
/usr/local/lib/python2.7/dist-packages/zipline-0.7.0-py2.7.egg/zipline/gens/tradesimulation.pyc in transform(self, stream_in)
181 self.algo.performance_needs_update = True
182
--> 183 risk_message = self.algo.perf_tracker.handle_simulation_end()
184 yield risk_message
185
/usr/local/lib/python2.7/dist-packages/zipline-0.7.0-py2.7.egg/zipline/finance/performance/tracker.pyc in handle_simulation_end(self)
466 ars,
467 self.sim_params,
--> 468 benchmark_returns=bms)
469
470 risk_dict = self.risk_report.to_dict()
/usr/local/lib/python2.7/dist-packages/zipline-0.7.0-py2.7.egg/zipline/finance/risk/report.pyc in __init__(self, algorithm_returns, sim_params, benchmark_returns)
83 end_date = self.algorithm_returns.index[-1]
84
---> 85 self.month_periods = self.periods_in_range(1, start_date, end_date)
86 self.three_month_periods = self.periods_in_range(3, start_date,
87 end_date)
/usr/local/lib/python2.7/dist-packages/zipline-0.7.0-py2.7.egg/zipline/finance/risk/report.pyc in periods_in_range(self, months_per, start, end)
132 end_date=cur_end,
133 returns=self.algorithm_returns,
--> 134 benchmark_returns=self.benchmark_returns
135 )
136
/usr/local/lib/python2.7/dist-packages/zipline-0.7.0-py2.7.egg/zipline/finance/risk/period.pyc in __init__(self, start_date, end_date, returns, benchmark_returns)
68 self.algorithm_returns = self.mask_returns_to_period(returns)
69 self.benchmark_returns = self.mask_returns_to_period(benchmark_returns)
---> 70 self.calculate_metrics()
71
72 def calculate_metrics(self):
/usr/local/lib/python2.7/dist-packages/zipline-0.7.0-py2.7.egg/zipline/finance/risk/period.pyc in calculate_metrics(self)
125 self.information = self.calculate_information()
126 self.beta, self.algorithm_covariance, self.benchmark_variance, \
--> 127 self.condition_number, self.eigen_values = self.calculate_beta()
128 self.alpha = self.calculate_alpha()
129 self.excess_return = self.algorithm_period_returns - \
/usr/local/lib/python2.7/dist-packages/zipline-0.7.0-py2.7.egg/zipline/finance/risk/period.pyc in calculate_beta(self)
254 self.benchmark_returns])
255 C = np.cov(returns_matrix, ddof=1)
--> 256 eigen_values = la.eigvals(C)
257 condition_number = max(eigen_values) / min(eigen_values)
258 algorithm_covariance = C[0][1]
/usr/lib/python2.7/dist-packages/numpy/linalg/linalg.pyc in eigvals(a)
886 _assertRankAtLeast2(a)
887 _assertNdSquareness(a)
--> 888 _assertFinite(a)
889 t, result_t = _commonType(a)
890
/usr/lib/python2.7/dist-packages/numpy/linalg/linalg.pyc in _assertFinite(*arrays)
215 for a in arrays:
216 if not (isfinite(a).all()):
--> 217 raise LinAlgError("Array must not contain infs or NaNs")
218
219 def _assertNoEmpty2d(*arrays):
LinAlgError: Array must not contain infs or NaNs
pdiffley
twiecki
mosesmc52
ssanderson
jetheurer
When running Zipline with ^BVSP benchmark (a simulation with start in 2008-4-1 and end in 2008-5-1), zipline crashs with the following error:
raise LinAlgError("Array must not contain infs or NaNs") numpy.linalg.linalg.LinAlgError: Array must not contain infs or NaNs
This is due a incosistence in Yahoo benchmark data. In Brazil, Tiradentes Holiday is in April 21, but this date is in Yahoo data as zero return. When running the algorithm, the function handle_market_close() misaligns his "todays_date" with all_benchmark_returns, and RiskMetricsCumulative finally gets "NaN" as valid benchmark returns.