def knn_test_several_params(trnX, tstX, trnY, tstY, multi_class, plot=False):
nvalues = [1, 3]
dist = ["manhattan", "euclidean", "chebyshev"]
values = {}
best_n_value = 0
best_dist = 0
best_score = 0
best_model = None
for d in dist:
yvalues = []
for n in nvalues:
knn = knn_model(trnX, trnY, n, d)
score = calculte_models_auc_score(knn, tstX, tstY, multi_class)
if score > best_score:
best_score = score
best_n_value = n
best_dist = d
best_model = knn
yvalues.append(score)
values[d] = yvalues
if plot:
plt.figure()
multiple_line_chart(plt.gca(),
nvalues,
values,
"KNN variants",
"n",
"Sensitivty",
percentage=True)
plt.show()
return best_model, best_score, best_dist, best_n_value
def xg_boost(trnX, tstX, trnY, tstY, multi_class, plot=False):
best_score = 0
best_model = None
best_learning_rate = 0
best_depth = 0
best_estimator = 0
learning_rates = [0.1, 0.05, 0.01]
max_depths = [5, 25]
n_estimators = [10, 50, 100, 200, 300, 400, 500]
if plot:
plt.figure()
fig, axs = plt.subplots(1,
len(max_depths),
figsize=(16, 4),
squeeze=False)
for k in range(len(max_depths)):
max_depth = max_depths[k]
values = {}
for learning_rate in learning_rates:
yvalues = []
for n in n_estimators:
model = xgb.XGBClassifier(
max_depth=max_depth,
learning_rate=learning_rate,
n_estimators=n,
early_stopping_rounds=10,
)
model.fit(trnX, trnY)
score = calculte_models_auc_score(model, tstX, tstY,
multi_class)
if score > best_score:
best_score = score
best_model = model
best_learning_rate = learning_rate
best_depth = max_depth
best_estimator = n
yvalues.append(score)
values[learning_rate] = yvalues
if plot:
multiple_line_chart(
axs[0, k],
n_estimators,
values,
"XG Boost with %s depth" % max_depth,
"Number of estimators",
"Sensitivity",
percentage=True,
)
if plot:
plt.show()
return best_model, best_score, best_learning_rate, best_depth, best_estimator
def dt_plot_accuracy(trnX, tstX, trnY, tstY, multi_class, plot=False):
best_samples_leaf = 0
best_depth = 0
best_criteria = ""
best_score = 0
best_model = None
min_samples_leaf = [.01]
max_depths = [10]
criteria = ["entropy", "gini"]
if plot:
plt.figure()
fig, axs = plt.subplots(1, 2, figsize=(16, 4), squeeze=False)
for k in range(len(criteria)):
f = criteria[k]
values = {}
for d in max_depths:
yvalues = []
for n in min_samples_leaf:
tree = decision_tree(trnX,
trnY,
samples_leaf=n,
depth=d,
criterion=f)
score = calculte_models_auc_score(tree, tstX, tstY,
multi_class)
if score > best_score:
best_score = score
best_depth = d
best_criteria = f
best_samples_leaf = n
best_model = tree
yvalues.append(score)
values[d] = yvalues
if plot:
multiple_line_chart(
axs[0, k],
min_samples_leaf,
values,
"%s criteria" % f,
"min sample leaf",
"Sensitivity",
percentage=True,
)
if plot:
plt.show()
return best_model, best_score, best_samples_leaf, best_depth, best_criteria
def rf_test_different_params(trnX, tstX, trnY, tstY, multi_class, plot=False):
best_numb_estimator = 0
best_depth = 0
best_feature = ""
best_score = 0
best_model = None
n_estimators = [100]
max_depths = [10]
max_features = ["sqrt", "log2"]
if plot:
plt.figure()
fig, axs = plt.subplots(1, 2, figsize=(10, 4), squeeze=False)
for k in range(len(max_features)):
f = max_features[k]
values = {}
for d in max_depths:
yvalues = []
for n in n_estimators:
rf = random_forest(trnX, trnY, n, d, f)
score = calculte_models_auc_score(rf, tstX, tstY, multi_class)
if score > best_score:
best_score = score
best_depth = d
best_numb_estimator = n
best_feature = f
best_model = rf
yvalues.append(score)
values[d] = yvalues
if plot:
multiple_line_chart(
axs[0, k],
n_estimators,
values,
"RF with %s features" % f,
"nr estimators",
"Sensitivity",
percentage=True,
)
if plot:
plt.show()
return best_model, best_score, best_numb_estimator, best_depth, best_feature
def histogram_with_distributions(ax: plt.Axes, series: pd.Series, var: str):
values = series.sort_values().values
n, bins, patches = ax.hist(values, 20, density=True, edgecolor="grey")
distributions = compute_known_distributions(values, bins)
multiple_line_chart(ax, values, distributions, "Best fit for %s" % var,
var, "probability")