def main(num_iters = 5):
pyodbc.lowercase = False
conn = pyodbc.connect(r"DRIVER=SQL Server;Trusted_Connection=Yes;DATABASE=pardee_datascience;SERVER=kpwhri_datascience.ghc.org")
conn.execute("insert into ml_runs (runtime) values (DEFAULT)")
conn.commit()
run_id = conn.execute("select max(run_id) as run_id from ml_runs").fetchone()[0]
xs, ys, acg_preds = fetch_data(n = 80000, connection = conn)
# print(acg_preds[0:3])
# print(len(acg_preds))
for i in range(num_iters):
# x_train, x_test, y_train, y_test = train_test_split(xs, ys,
# test_size=0.33,
# random_state=0,
# stratify=ys)
x_train, x_test, y_train, y_test = dev_val_split(xs, ys)
preds = y_test.copy(deep=True)
# preds = pd.DataFrame({'hospitalized': y_test})
print_stats(y_train, 'Training')
print_stats(y_test, 'Test')
clf = Perceptron(max_iter=1000, tol = .001) # TODO: find out what these params do (apart from suppress a warning)
clf.nickname = "perceptron "
try_pred(run_id, conn, clf, y_train, y_test, x_train, x_test)
clf = LinearSVC(random_state=i)
clf.nickname = "LinearSVC"
try_pred(run_id, conn, clf, y_train, y_test, x_train, x_test)
clf = RandomForestClassifier(n_jobs=-1, max_leaf_nodes=100, random_state=i)
clf.nickname = "RandmForest"
preds["RandmForest"] = try_pred(run_id, conn, clf, y_train, y_test, x_train, x_test)
clf = GaussianNB()
clf.nickname = "naive bayes"
preds["naive_bayes"] = try_pred(run_id, conn, clf, y_train, y_test, x_train, x_test)
# clf = LinearDiscriminantAnalysis(solver='svd')
# clf.nickname = "LDA "
# preds["LDA"] = try_pred(run_id, conn, clf, y_train, y_test, x_train, x_test)
clf = svm.SVC(probability=True)
clf.nickname = "SVM "
preds["SVM"] = try_pred(run_id, conn, clf, y_train, y_test, x_train, x_test)
preds["Random"] = np.random.rand(preds.shape[0])
preds = pd.merge(preds, acg_preds, left_index=True, right_index=True)
preds.to_csv(HOME_DIR + "preds_d.csv")
record_predictions(run_id, conn, preds)
conn.close()
