def analyze_weights(model: Model, columns):
zero_sample = pd.DataFrame(np.zeros((1, len(columns))), columns=columns)
offset = model.predict(zero_sample).values.ravel()
weights = pd.DataFrame(offset, columns=["offset"])
for column in columns:
zero_sample[column] = 1.0
weights[column] = model.predict(zero_sample)
weights[column] -= weights["offset"]
zero_sample[column] = 0.0
return weights
def evaluateSVM(settings, grid={}):
svmModel = Model("train_and_test",
features=[
"bow", "kuhn_munkres", "length_diff", "q_features",
"ref_hedg_bow", "root_dist", "SVO_ppdb", "word2vec"
],
classifier="SVM",
settings=settings,
hyperparameters_grid=grid)
print("Results SVM: ", svmModel.results)
return svmModel
def evaluateNaiveBayes(settings):
naiveBayesModel = Model("train_and_test",
features=[
"bow", "kuhn_munkres", "length_diff",
"q_features", "ref_hedg_bow", "root_dist",
"SVO_ppdb", "word2vec"
],
classifier="Naive Bayes",
settings=settings)
print("Results Naïve Bayes: ", naiveBayesModel.results)
return naiveBayesModel
def evaluateLogisticRegression(settings):
logisticRegressionModel = Model("train_and_test",
features=[
"bow", "kuhn_munkres", "length_diff",
"q_features", "ref_hedg_bow",
"root_dist", "SVO_ppdb", "word2vec"
],
classifier="Logistic Regression",
settings=settings)
print("Results Logistic Regression: ", logisticRegressionModel.results)
return logisticRegressionModel
target_transformations = [
ColumnSelection([volume_col]),
CutNTop(horizon)
]
transformations = DataTransformations(feature_transformations, target_transformations)
splitter = Splitter(ratio=0.2)
dataset_maker = DatasetMaker(transformations, splitter)
# Data
data = pd.read_csv("data/sp500.csv")
training_data, testing_data = dataset_maker.make(data)
# Learning
params = {
"solver": "adam",
"activation": "tanh",
"hidden_layer_sizes": tuple(20 for _ in range(5)),
"max_iter": 10000,
}
model = Model(MLPRegressor, params)
model.fit(*training_data)
prediction = model.predict(testing_data.features)
pprint(model.get_parameters())
BarsDrawer(testing_data.target, prediction).show()
print("R2:", r2_score(testing_data.target, prediction))
transformations = DataTransformations(feature_transformations, target_transformations)
splitter = Splitter(ratio=0.4, seed=42)
dataset_maker = DatasetMaker(transformations, splitter)
# Data
data = pd.read_csv("data/sp500.csv")
training_data, testing_data = dataset_maker.make(data)
# Learning
params = {
"fit_intercept": False,
}
model = Model(LinearRegression, params)
model.fit(*training_data)
prediction = model.predict(testing_data.features)
BarsDrawer(testing_data.target, prediction).show()
print("Coefficients:", model.model.coef_)
pprint(model.get_parameters())
print("R2:", r2_score(testing_data.target, prediction))
print("mean abs error:", mean_absolute_error(testing_data.target, prediction))
print("% mean abs error:", relative_absolute_error_per_sample(testing_data.target, prediction))
columns = testing_data.features.columns
zero_sample = pd.DataFrame(np.zeros((1, len(columns))), columns=columns)
trainingSettingsSVM, trainingSettingsRandomForest, nestedRandomForestGrid, nestedRandomForestSettings, nestedSVMSettings
import json
feature_set = {
"bow", "kuhn_munkres", "length_diff", "q_features", "ref_hedg_bow",
"root_dist", "SVO_ppdb", "word2vec"
}
classifiers = [("Logistic Regression", trainingSettingsLogisticRegression, {}),
("Naive Bayes", trainingSettingsNaiveBayes, {}),
("Random Forest", nestedRandomForestSettings,
nestedRandomForestGrid),
("SVM", trainingSettingsSVM, nestedSVMSettings)]
results = {}
for feature in feature_set:
new_feature_set = feature_set.copy()
new_feature_set.remove(feature)
results[feature] = {}
for (clsf, settings, hyper) in classifiers:
print(
f'Fitting model {clsf} with parameters {settings} and hyper {hyper}'
)
model = Model("train_and_test",
features=new_feature_set,
classifier=clsf,
settings=settings,
hyperparameters_grid=hyper)
results[feature][clsf] = model.results
with open(DATA_PATH + "/ablation_results.json", "w") as fp:
json.dump(results, fp, indent=4)