def test_simple_benchmark():
sns.set_style("whitegrid")
data_obj = load_breast_cancer()
x = data_obj['data']
y = data_obj['target']
names = data_obj['feature_names']
# let's overfit, just for demo purposes
clf = san.SAN(num_epochs=64,
num_heads=2,
batch_size=8,
dropout=0.2,
hidden_layer_size=32,
learning_rate=0.01)
x = sparse.csr_matrix(x)
clf.fit(x, y)
predictions = clf.predict(x)
global_attention_weights = clf.get_mean_attention_weights()
local_attention_matrix = clf.get_instance_attention(x)
mutual_information = mutual_info_classif(x, y)
rf_model = RandomForestClassifier()
rf_model.fit(x, y)
rf = rf_model.feature_importances_
sorted_rf = np.argsort(rf)
sorted_mi = np.argsort(mutual_information)
sorted_local_attention = np.argsort(np.max(local_attention_matrix, axis=0))
sorted_global_attention = np.argsort(global_attention_weights)
names = ["RF", "MI", "attention-local", "attention-global"]
indices = [
sorted_rf, sorted_mi, sorted_local_attention, sorted_global_attention
]
output_struct = {}
for name, indice_set in zip(names, indices):
scores = []
indice_set = indice_set.tolist()
print("Computing evaluations for: {}".format(name))
for j in tqdm.tqdm(range(len(indice_set))):
selected_features = indice_set[0:j + 1]
subset = x[:, selected_features]
clf = LogisticRegression(max_iter=10000000, solver="lbfgs")
score = np.mean(cross_val_score(clf, subset, y, cv=10))
scores.append((score, j + 1))
output_struct[name] = scores
print("Plotting ..")
for k, v in output_struct.items():
indices = []
scores = []
for x, y in v:
indices.append(y)
scores.append(x)
plt.plot(indices, scores, label=k)
plt.xlabel("Top features")
plt.ylabel("Performance (Accuracy)")
plt.legend()
plt.show()
def test_simple_ranking():
sns.set_style("whitegrid")
dataobj = load_breast_cancer()
x = dataobj['data']
y = dataobj['target']
# let's overfit, just for demo purposes
clf = san.SAN(num_epochs=18, num_head=2, batch_size=8, dropout=0.05, hidden_layer_size=32)
x = sparse.csr_matrix(x)
clf.fit(x, y)
predictions = clf.predict(x)
global_attention_weights = clf.get_mean_attention_weights()
local_attention_matrix = clf.get_instance_attention(x.todense())
return predictions, global_attention_weights, local_attention_matrix
def test_simple_ranking():
sns.set_style("whitegrid")
data_obj = load_breast_cancer()
x = data_obj['data']
y = data_obj['target']
names = data_obj['feature_names']
# let's overfit, just for demo purposes
clf = san.SAN(num_epochs=32,
num_heads=2,
batch_size=8,
dropout=0.2,
hidden_layer_size=32)
x = sparse.csr_matrix(x)
clf.fit(x, y)
predictions = clf.predict(x)
global_attention_weights = clf.get_mean_attention_weights()
local_attention_matrix = clf.get_instance_attention(x.todense())
mutual_information = mutual_info_classif(x, y)
rf_model = RandomForestClassifier()
rf_model.fit(x, y)
rf = rf_model.feature_importances_
plt.plot(names,
global_attention_weights,
label="Global attention",
marker="x")
plt.plot(names,
np.mean(local_attention_matrix, axis=0),
label="Local attention - mean",
marker="x")
plt.plot(names,
np.max(local_attention_matrix, axis=0),
label="Local attention - max",
marker="x")
plt.plot(names, mutual_information, label="Mutual information", marker=".")
plt.plot(names, rf, label="RandomForest", marker=".")
plt.legend(loc=1)
plt.xticks(rotation=90)
plt.tight_layout()
plt.show()
return predictions
from sklearn.feature_selection import mutual_info_classif # chi2, f_classif
from sklearn.ensemble import RandomForestClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import cross_val_score
import tqdm
from sklearn.model_selection import KFold
data_obj = load_iris()
x = data_obj['data']
y = data_obj['target']
names = data_obj['feature_names']
clf = san.SAN(num_epochs=30,
num_heads=4,
batch_size=16,
dropout=0.2,
hidden_layer_size=32,
stopping_crit=10,
learning_rate=0.01)
kf = KFold(n_splits=10)
accuracy_results = []
for train_index, test_index in kf.split(x):
train_x = x[train_index]
test_x = x[test_index]
train_y = y[train_index]
test_y = y[test_index]
x_sp = sparse.csr_matrix(train_x)
xt_sp = sparse.csr_matrix(test_x)
clf.fit(x_sp, train_y)
predictions = clf.predict(xt_sp)