def feature_extraction(data_df, inputs):
"""
feature extraction --> Transform the existing features
into a lower dimensional space
Transforms the data - can be used to linearly separate
data thru dimensionality reduction
"""
timeme(principal_component_analysis)(data_df, tuple(inputs))
def feature_selection(train_df, inputs):
"""
Sequential Backward Selection - feature selection to see
which are the most telling variable
Default is K-means Clustering
Feature selection: Select a subset of the existing
features without a transformation
Use this to limit down the factors we learn on
"""
ests = []
ests.append([
DecisionTreeClassifier(criterion='entropy',
max_depth=3,
random_state=0), 'DecTree'
])
ests.append([
RandomForestClassifier(criterion='entropy',
n_estimators=3,
random_state=1,
n_jobs=3), 'RandForest'
])
# ests.append([SVC(kernel='linear', C=100, random_state=0), 'SVC'])
ests.append(
[LogisticRegression(C=100, random_state=0, penalty='l1'), 'LogRegr'])
# ests.append([AdalineSGD(n_iter=15, eta=0.001, random_state=1),
# 'AdalineSGD'])
# ests.append([AdalineGD(n_iter=20, eta=0.001), 'AdalineGD'])
ests.append([KNeighborsClassifier(n_neighbors=3), 'Kmeans'])
ranks = []
# for ind_est in ests:
# print("running for {}".format(ind_est[1]))
# ranks.append([ind_est[1], timeme(sbs_run)(train_df, tuple(inputs),
# est=ind_est[0], name=ind_est[1])])
# Random Forest Feature Selection - using a random forest to identify
# which factors decrease impurity the most
pdb.set_trace()
ranks.append([
timeme(random_forest_feature_importance)(train_df, tuple(inputs)),
'RandForestFeats'
])
# Logistic Regression Feature Selection - logistic regression
# should expose the important variables through its weights
pdb.set_trace()
ranks.append([
timeme(logistic_regression_feature_importance)(train_df,
tuple(inputs)),
"LogRegrWgts"
])
pdb.set_trace()
for rank in ranks:
print("Ranks for {}".format(rank[1]))
print(rank[0])
def run_analysis():
"""
Function to run thru the whole analysis process
"""
# Run the filter
print("Running Screener:")
ticks = timeme(run_filter)()
print("{} stocks thru filter\n".format(len(ticks)))
# check recent momentum returns
print("Running Momentum Check:")
ticks = timeme(check_momentum)('20190703', ticks)
print("{} stocks thru momentum checks\n".format(len(ticks)))
# check recent big vs small results
print("Running Big vs. Small Filter:")
ticks = timeme(check_big_v_small)('20190703', ticks.reset_index())
print("{} stocks thru big vs small filter\n".format(len(ticks)))
# remove ticks that should be ignored
print("Ignoring Certain Symbols:")
ticks = timeme(ignore_ticks)(ticks.reset_index())
print("{} stocks thru ignore filter\n".format(len(ticks)))
# Run equity valuation
print("Running Equity Valuation on: {}".format(
ticks.index.levels[0].values))
timeme(run_eq_valuation)(ticks)
print()
def model_evaluation(data_df, inputs):
"""
Evaluate the performance of your model thru different techniques
"""
timeme(kfold_cross_validation)(data_df, tuple(inputs))
def model_evaluation(df, inputs):
timeme(kfold_cross_validation)(df, tuple(inputs))