def fitness(self, organism):
cache_key = organism.__repr__()
if self.cache.has_key(cache_key):
print "cache hit", cache_key
sc, test_pred = self.cache[cache_key]
organism.found_forest = FoundForest(cache_key, sc, test_pred)
print "cache score", sc
return sc
def feature_generator(__sp):
return TechnicalFeatures(self.stock_period, feature_desc=organism.features_hash())
forest_options = False
# if organism["n_estimators"]:
# max = None
# if organism["max_features"] != "None":
# max = organism["max_features"]
# forest_options = {"n_estimators": organism["n_estimators"],
# "min_samples_split": organism["min_samples_split"],
# "max_features": max
# }
mlearn = MachineLearner(
self.settings.symbol,
classifier=cl.BuySellClassifier(threshold=organism["threshold"], days=organism["days_return"]),
feature_gen=feature_generator,
stock_period=self.stock_period,
random_forest_options=forest_options,
)
mlearn.learn_period(self.settings.learn_start_time, self.settings.learn_end_time)
cache_predictions = mlearn.predict_period(
self.settings.predict_cache_start_time, self.settings.predict_cache_end_time
)
print organism
score = getattr(FitnessScoring, self.settings.fitness_method)(
mlearn, cache_predictions, self.settings.val_start_time, self.settings.val_end_time
)
# cache_test_predictions = False
# if self.settings.precache_tests():
# cache_test_predictions = mlearn.predict_period(self.settings.test_start_time, self.settings.test_end_time)
organism.found_forest = FoundForest(cache_key, score, cache_predictions)
self.history.append(organism.copy())
self.cache[cache_key] = (score, cache_predictions)
return score
def setUp(self):
self.start_date = dt.datetime(2004,9,1)
self.end_date = dt.datetime(2010,1,1)
self.sp_limited = StockPeriod('GOOG_SHORT', self.start_date, self.end_date)
self.tf_limited = TechnicalFeatures(self.sp_limited)
self.lim_start = dt.datetime(2004,9,1,16)
self.lim_end = dt.datetime(2009,12,31,16)
self.mlearn_lim = MachineLearner('GOOG_SHORT', feature_gen = lambda sp: self.tf_limited, stock_period = self.sp_limited)
self.sp = StockPeriod('GOOG', self.start_date, dt.datetime(2011,1,1))
self.tf = TechnicalFeatures(self.sp)
self.mlearn_verify = MachineLearner('GOOG_', feature_gen = lambda sp: self.tf, stock_period = self.sp)
class TestVerifyThatPredictionsArePrecient(unittest.TestCase):
def setUp(self):
self.start_date = dt.datetime(2004,9,1)
self.end_date = dt.datetime(2010,1,1)
self.sp_limited = StockPeriod('GOOG_SHORT', self.start_date, self.end_date)
self.tf_limited = TechnicalFeatures(self.sp_limited)
self.lim_start = dt.datetime(2004,9,1,16)
self.lim_end = dt.datetime(2009,12,31,16)
self.mlearn_lim = MachineLearner('GOOG_SHORT', feature_gen = lambda sp: self.tf_limited, stock_period = self.sp_limited)
self.sp = StockPeriod('GOOG', self.start_date, dt.datetime(2011,1,1))
self.tf = TechnicalFeatures(self.sp)
self.mlearn_verify = MachineLearner('GOOG_', feature_gen = lambda sp: self.tf, stock_period = self.sp)
def assert_limits(self, sp, tf, start_d, end_d):
self.assertEqual(sp.close_data.index[0], start_d)
self.assertEqual(sp.close_data.index[-1], end_d)
self.assertEqual(tf.relative_data['close'].index[0], start_d)
self.assertEqual(tf.relative_data['close'].index[-1], end_d)
def test_correct_boundaries(self):
self.assert_limits(self.sp_limited, self.tf_limited, self.lim_start, self.lim_end)
self.assert_limits(self.mlearn_lim.sp, self.mlearn_lim.feats, self.lim_start, self.lim_end)
def test_predictions_are_the_same(self):
#predictions after end date
features = self.tf.get_features()[:][self.lim_end:][1:100]
self.assertTrue(features.index[0] > self.lim_end)
self.mlearn_lim.learn_period(dt.datetime(2006,1,1), dt.datetime(2008,1,1))
self.assert_limits(self.mlearn_lim.sp, self.mlearn_lim.feats, self.lim_start, self.lim_end)
data_holder = []
for index, feat_row in features.iterrows():
res = self.mlearn_lim.predict([feat_row.values])
data_holder.append(res[0])
df_test_result = pd.DataFrame(index = features.index, data = data_holder)
print df_test_result
self.mlearn_verify.learn_period(dt.datetime(2006,1,1), dt.datetime(2008,1,1))
df_verify_result = self.mlearn_verify.predict_period(features.index[0], features.index[-1])
print df_verify_result
self.assertTrue(np.equal(df_test_result.index, df_verify_result.index).all())
self.assertTrue(np.equal(df_test_result.values, df_verify_result.values).all())
def test_features_equal(self):
limited_feats = self.tf_limited.get_features()
feats = self.tf.get_features()
sub_set_limited = limited_feats[:][-50:]
#print sub_set_limited
self.assertEqual(sub_set_limited.index[-1], self.lim_end)
sub_set_full = feats[:][sub_set_limited.index[0]:sub_set_limited.index[-1]]
#print sub_set_full
self.assertTrue(feats.index[-1] > self.lim_end)
#print np.equal(sub_set_limited.values, sub_set_full.values)
self.assertTrue(np.equal(sub_set_limited.values, sub_set_full.values).all())
