def produce_bagging_models(prices, numbags=11, minority=1, batch_size=21, **preprocess_params):
# preprocess the training data
prices_vectors, wrangled_annotations = pqp.preprocess_trainingdata(prices, **preprocess_params)
# separate into minority and majority
minority_pairs = filter(lambda item: item[1]==minority, zip(prices_vectors, wrangled_annotations))
majority_pairs = filter(lambda item: item[1]!=minority, zip(prices_vectors, wrangled_annotations))
# sampling and bagging
window_size = preprocess_params['window_size'] if preprocess_params.has_key('window_size') else pqp.def_prediction_params['window_size']
len_minority = len(minority_pairs)
len_majority = len(majority_pairs)
models = []
choices = []
for i in range(numbags):
selected_majidx = np.random.randint(0, len_majority, len_minority)
sampled_pairs = minority_pairs + map(lambda i: majority_pairs[i], selected_majidx)
sampled_prvecs = np.array(map(lambda pair: pair[0], sampled_pairs))
sampled_wranns = np.array(map(lambda pair: pair[1], sampled_pairs))
model = pqp.train_prediction_model_int(sampled_prvecs, sampled_wranns,
window_size=window_size,
batch_size=batch_size)
models.append(model)
choices.append(selected_majidx)
# return
return {'models': models, 'selected_indices': choices}
def batch_pricequenchmodels_prediction(prices, models,
window_size=pqp.def_prediction_params['window_size'],
future_window=pqp.def_annotation_params['future_window'],
to_normalize=True
):
prices_vectors = pqp.wrangling_pricevector(prices,
window_size=window_size,
future_window=future_window,
to_normalize=to_normalize)
prediction_matrix = np.array(map(lambda model: map(lambda price: pqp.pricequench_predict(price, model), prices_vectors), models))
prediction_matrix = np.reshape(prediction_matrix, prediction_matrix.shape[:2])
return prediction_matrix
def batch_pricequenchmodels_prediction(
prices,
models,
window_size=pqp.def_prediction_params['window_size'],
future_window=pqp.def_annotation_params['future_window'],
to_normalize=True):
prices_vectors = pqp.wrangling_pricevector(prices,
window_size=window_size,
future_window=future_window,
to_normalize=to_normalize)
prediction_matrix = np.array(
map(
lambda model: map(
lambda price: pqp.pricequench_predict(price, model),
prices_vectors), models))
prediction_matrix = np.reshape(prediction_matrix,
prediction_matrix.shape[:2])
return prediction_matrix
def produce_bagging_models(prices,
numbags=11,
minority=1,
batch_size=21,
**preprocess_params):
# preprocess the training data
prices_vectors, wrangled_annotations = pqp.preprocess_trainingdata(
prices, **preprocess_params)
# separate into minority and majority
minority_pairs = filter(lambda item: item[1] == minority,
zip(prices_vectors, wrangled_annotations))
majority_pairs = filter(lambda item: item[1] != minority,
zip(prices_vectors, wrangled_annotations))
# sampling and bagging
window_size = preprocess_params['window_size'] if preprocess_params.has_key(
'window_size') else pqp.def_prediction_params['window_size']
len_minority = len(minority_pairs)
len_majority = len(majority_pairs)
models = []
choices = []
for i in range(numbags):
selected_majidx = np.random.randint(0, len_majority, len_minority)
sampled_pairs = minority_pairs + map(lambda i: majority_pairs[i],
selected_majidx)
sampled_prvecs = np.array(map(lambda pair: pair[0], sampled_pairs))
sampled_wranns = np.array(map(lambda pair: pair[1], sampled_pairs))
model = pqp.train_prediction_model_int(sampled_prvecs,
sampled_wranns,
window_size=window_size,
batch_size=batch_size)
models.append(model)
choices.append(selected_majidx)
# return
return {'models': models, 'selected_indices': choices}
