def test_iRF_weight1():
# Check when label is random, whether the feature importance of every
# feature is the same.
n_samples = 1000
n_features = 10
random_state_classifier = 2018
np.random.seed(random_state_classifier)
X_train = np.random.uniform(low=0, high=1, size=(n_samples, n_features))
y_train = np.random.choice([0, 1], size=(n_samples, ), p=[.5, .5])
X_test = np.random.uniform(low=0, high=1, size=(n_samples, n_features))
y_test = np.random.choice([0, 1], size=(n_samples, ), p=[.5, .5])
all_rf_weights, all_K_iter_rf_data, \
all_rf_bootstrap_output, all_rit_bootstrap_output, \
stability_score = irf_utils.run_iRF(X_train=X_train,
X_test=X_test,
y_train=y_train,
y_test=y_test,
K=5,
n_estimators=20,
B=30,
random_state_classifier=2018,
propn_n_samples=.2,
bin_class_type=1,
M=20,
max_depth=5,
noisy_split=False,
num_splits=2,
n_estimators_bootstrap=5)
assert np.max(all_rf_weights['rf_weight5']) < .135
def test_iRF_weight2():
# Check when feature 1 fully predict the label, its importance should be 1.
n_samples = 1000
n_features = 10
random_state_classifier = 2018
np.random.seed(random_state_classifier)
X_train = np.random.uniform(low=0, high=1, size=(n_samples, n_features))
y_train = np.random.choice([0, 1], size=(n_samples, ), p=[.5, .5])
X_test = np.random.uniform(low=0, high=1, size=(n_samples, n_features))
y_test = np.random.choice([0, 1], size=(n_samples, ), p=[.5, .5])
# first feature is very important
X_train[:, 1] = X_train[:, 1] + y_train
X_test[:, 1] = X_test[:, 1] + y_test
all_rf_weights, all_K_iter_rf_data, \
all_rf_bootstrap_output, all_rit_bootstrap_output, \
stability_score = irf_utils.run_iRF(X_train=X_train,
X_test=X_test,
y_train=y_train,
y_test=y_test,
K=5,
n_estimators=20,
B=30,
random_state_classifier=2018,
propn_n_samples=.2,
bin_class_type=1,
M=20,
max_depth=5,
noisy_split=False,
num_splits=2,
n_estimators_bootstrap=5)
print(all_rf_weights['rf_weight5'])
assert all_rf_weights['rf_weight5'][1] == 1
# prep
X = dmx[1:, 1:].astype(float).T
# assigning a nurmerical variable to the string classifiers
# 0 for cntrol, 1 for compared
# biclass
y = np.array(itemgetter(*idx[1:, 2])({'Control': 0, 'D12': 1}))
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)
# iRF
irfres = irf_utils.run_iRF(
X_train=X_train,
X_test=X_test,
y_train=y_train,
y_test=y_test,
rf=RandomForestClassifierWithWeights(n_estimators=100),
K=5, # number of iteration. This is recommended value by dev
B=100, # The number of bootstrap samples. Play around with this, see what changes
M=137, # number of trees (RIT) to build. Look into the effects of this parameter
max_depth=5,
)
rf_weights, K_iter_rf_data, rf_bootstrap_output, rit_bootstrap_output, stability_score = irfres
# feature importance
# i probably don't need to worry about the rest of this, i.e modifying etc
fids = dmx[1:, 0]
iteration = 'rf_iter5'
impt = K_iter_rf_data[iteration]['feature_importances']
impt_std = K_iter_rf_data[iteration]['feature_importances_std']
#X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.3)
# trying a better version of testtrain split
kf = KFold(n_splits=5, random_state=15, shuffle=True)
for count_k, (train_index, test_index) in enumerate(kf.split(X)):
X_train = X[train_index]
X_test = X[test_index]
y_train = y[train_index]
y_test = y[test_index]
irfres = irf_utils.run_iRF(
X_train=X_train,
X_test=X_test,
y_train=y_train,
y_test=y_test,
rf=RandomForestClassifierWithWeights(n_estimators=30),
K=10, # number of iteration
B=30, # The number of bootstrap samples
M=20, # number of trees (RIT) to build
max_depth=5,
)
rf_weights, K_iter_rf_data, rf_bootstrap_output, rit_bootstrap_output, stability_score = irfres
# feature importance
fids = dmx[1:, 0]
iteration = 'rf_iter5'
impt = K_iter_rf_data[iteration]['feature_importances']
impt_std = K_iter_rf_data[iteration]['feature_importances_std']
impt_rank_idx = K_iter_rf_data[iteration]['feature_importances_rank_idx']