def day_offset_sampler(window=1):
"""Assuming observations from dataset are (x, y), compute the density of the y's over the next n days
Args:
window (int): how many days to use in the density
Yield:
[(x₁, y₁), (x₂, y₂), ...)]: where ∀xᵢ are part of the same day, and y is a density over the next {offset} days
"""
buffer = deque()
sampler = dataset_sampler(x_format='OneHot', y_format='Ordinal')
def add_day(day):
# indexed: [newest day][first observation][first element of train pair][date index]
if len(buffer) and buffer[-1][0][0][0] == day[0][0]:
buffer[-1].append(day)
else:
buffer.append([day])
while len(buffer) <= window + 1:
add_day(next(sampler))
for observation in sampler:
add_day(observation)
if len(buffer) > window + 1:
current_day = buffer.popleft()
temp = buffer.pop(
) # the last day only has one element outside the window
expected = np.mean([j[1] for i in buffer for j in i], axis=0)
buffer.append(temp)
yield [(record[0], expected) for record in current_day]
def run_simple():
from models.torch_simple.train import train_simple, test_simple
sampler = lambda: dataset_sampler(x_format='OneHot', y_format='Ordinal')
params = {'input_size': 25, 'output_size': 4}
train_simple(sampler, params)
evaluate(*test_simple(sampler, params))
def run_ann():
from models.torch_ann.train import train_ann, test_ann
sampler = lambda: dataset_sampler(x_format='OneHot', y_format='Ordinal')
params = {'input_size': 25, 'output_size': 4, 'layer_sizes': (100, 20)}
train_ann(sampler, params)
evaluate(*test_ann(sampler, params))
def day_sampler():
"""Assuming observations from dataset are (x, y), group by the day"""
sampler = dataset_sampler(x_format='OneHot', y_format='Ordinal')
buffer = [next(sampler)]
for observation in sampler:
if buffer[0][0][0] != observation[0][0]:
yield buffer
buffer = [observation]
buffer.append(observation)
yield buffer
def run_lstm():
from models.torch_lstm.train import train_lstm, test_lstm
sampler = lambda: dataset_sampler(x_format='OneHot', y_format='Ordinal')
params = {
'input_size': 25,
'output_size': 4,
'lstm_hidden_dim': 20,
'lstm_layers': 2,
'batch_size': 1
}
train_lstm(sampler, params)
evaluate(*test_lstm(sampler, params))
from sklearn.svm import SVC
from sklearn.naive_bayes import GaussianNB
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.ensemble import BaggingClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import AdaBoostClassifier
from sklearn.ensemble import GradientBoostingClassifier
model_specifications = [{
"name":
"Decision Tree",
"class":
DecisionTreeClassifier,
"datasource":
lambda: dataset_sampler(x_format='OneHot', y_format='Ordinal'),
"hyperparameters": {
"criterion": ["gini", "entropy"],
"splitter": ["best", "random"],
"min_samples_split": [2, 3],
"min_samples_leaf": [1, 5]
}
}, {
"name":
"Neural Network PCA 10pc",
"class":
MLPClassifier,
"datasource":
lambda: dataset_sampler(
x_format='OneHot', y_format='Ordinal', components=10),
"hyperparameters": {
def run_keras():
from models.keras_ann.network import train_keras, test_keras
sampler = lambda: dataset_sampler(x_format='OneHot', y_format='OneHot')
trainparams = {'epochs': 200, 'batch_size': 10}
train_keras(sampler, trainparams)
test_keras(sampler)