def batch_grid(kernel, wine):
if (kernel == 'rbf'):
C_range = np.logspace(0, 4, 5)
gamma_range = np.logspace(-2, 2, 5)
param_grid = [{'C': C_range, 'gamma': gamma_range, 'kernel': ['rbf']}]
elif (kernel == 'poly'):
C_range = np.logspace(-1, 3, 5)
gamma_range = np.logspace(-2, 2, 5)
param_grid = [{
'C': C_range,
'gamma': gamma_range,
'kernel': ['poly'],
'degree': [4, 5, 6],
'coef0': [0, 1, 2, 3, 4]
}]
elif (kernel == 'sigmoid'):
C_range = np.logspace(3, 7, 5)
gamma_range = np.logspace(-3, 1, 5)
param_grid = [{
'C': C_range,
'gamma': gamma_range,
'kernel': ['sigmoid'],
'coef0': [-1, 0, 1, 2, 3]
}]
df = u.get_wine(wine)
X_tr_norm, Y_tr, X_te_norm, Y_te = prepare_data(df)
grid_search = model_selection.GridSearchCV(
svm.SVR(epsilon=0.2, verbose=False, max_iter=100000),
param_grid,
n_jobs=2,
verbose=2,
scoring='neg_mean_absolute_error',
return_train_score=True,
cv=5)
t0 = time()
grid_search.fit(X_tr_norm, Y_tr)
print("done in %0.3fs" % (time() - t0))
print("Best score: %0.3f" % grid_search.best_score_)
print("Best parameters set:")
best_parameters = grid_search.best_estimator_.get_params()
print(best_parameters)
u.savegrid(grid_search, kernel, wine)
return grid_search
def grid_search_batch_size(prev_grid, wine):
mlp = MLPRegressor(**prev_grid.best_estimator_.get_params())
X_train, Y_train, X_test, Y_test = prepare_data(u.get_wine(wine))
params_grid = {}
grid_search = model_selection.GridSearchCV(mlp, params_grid, return_train_score=True, cv=5,
scoring='neg_mean_absolute_error', n_jobs=-1, verbose=5)
t0 = time()
grid_search.fit(X_train, Y_train)
u.savegrid(grid_search, "nn", wine, "batch_size")
print("done in %0.3fs" % (time() - t0))
print()
print("Best score: %0.3f" % grid_search.best_score_)
print("Best parameters set:")
best_parameters = grid_search.best_estimator_.get_params()
print(best_parameters)
return grid_search
def grid_search_alpha_lr(params, wine):
params_grid = {'alpha': np.logspace(-5, -3, 100), 'learning_rate_init': np.linspace(0.1, 1, 10)}
mlp = MLPRegressor(**params)
X_train, Y_train, X_test, Y_test = prepare_data(u.get_wine(wine))
grid_search = model_selection.GridSearchCV(mlp, params_grid, return_train_score=True, cv=5,
scoring='neg_mean_absolute_error', n_jobs=-1,
verbose=1)
t0 = time()
grid_search.fit(X_train, Y_train)
u.savegrid(grid_search, "nn", wine, "alpha_lr")
print("done in %0.3fs" % (time() - t0))
print()
print("Best score: %0.3f" % grid_search.best_score_)
print("Best parameters set:")
best_parameters = grid_search.best_estimator_.get_params()
print(best_parameters)
return grid_search
def grid_search_layers(grid_search, wine):
min_1 = 10
min_2 = 10
max_1 = 20
max_2 = 20
tuples = []
for l1 in np.linspace(min_1, max_1, 10, dtype=int):
for l2 in np.linspace(min_2, max_2, 10, dtype=int):
if l2 == 0:
tuples.append((l1,))
else:
tuples.append((l1, l2))
#layers_grid = [{'hidden_layer_sizes': tuples, 'activation': ['logistic', 'tanh', 'relu']}]
layers_grid = [{'hidden_layer_sizes': tuples, 'activation': ['relu']}]
mlp = MLPRegressor(max_iter=10000, solver='adam', tol=1e-4, random_state=999)
if (grid_search == None):
grid_search = model_selection.GridSearchCV(mlp, layers_grid, return_train_score=True, cv=5,
scoring='neg_mean_absolute_error', n_jobs=-1, verbose=3)
else:
grid_search.set_params(**layers_grid)
X_train, Y_train, X_test, Y_test = prepare_data(u.get_wine(wine))
t0 = time()
grid_search.fit(X_train, Y_train)
u.savegrid(grid_search, 'nn', wine, 'activation_layers')
print("done in %0.3fs" % (time() - t0))
print()
print("Best score: %0.3f" % grid_search.best_score_)
print("Best parameters set:")
best_parameters = grid_search.best_estimator_.get_params()
print(best_parameters)
return grid_search
def grid_searh_NN(wine,model_setup,param_grid,max_epochs =1000,n_cores=-1,verbose=0) :
mlp = MLPRegressor(**model_setup)
mlp.set_params(max_iter=max_epochs)
X_train, Y_train, X_test, Y_test = prepare_data(u.get_wine(wine))
grid_search = model_selection.GridSearchCV(mlp, param_grid, return_train_score=True, cv=5,
scoring='neg_mean_absolute_error', n_jobs=n_cores, verbose=verbose)
t0 = time()
grid_search.fit(X_train, Y_train)
f_path = ""
for i in param_grid.keys() :
f_path += i + "_"
u.savegrid(grid_search, "nn", wine, f_path)
print("done in %0.3fs" % (time() - t0))
print()
print("Best score: %0.3f" % grid_search.best_score_)
print("Best parameters set:")
best_parameters = grid_search.best_estimator_.get_params()
print(best_parameters)
return grid_search
def grid_search_beta(prev_grid, wine):
prev_params = prev_grid.best_estimator_.get_params()
prev_params['max_iter'] = 10000
mlp = MLPRegressor(**prev_params)
X_train, Y_train, X_test, Y_test = prepare_data(u.get_wine(wine))
params_grid = {'beta_1': np.linspace(0.5, 0.9999999999, 20), 'beta_2': np.linspace(0.51, 0.99999999999, 20)}
grid_search = model_selection.GridSearchCV(mlp, params_grid, return_train_score=True, cv=5,
scoring='neg_mean_absolute_error', n_jobs=-1,
verbose=5)
t0 = time()
grid_search.fit(X_train, Y_train)
u.savegrid(grid_search, "nn", wine, "beta")
print("done in %0.3fs" % (time() - t0))
print()
print("Best score: %0.3f" % grid_search.best_score_)
print("Best parameters set:")
best_parameters = grid_search.best_estimator_.get_params()
print(best_parameters)
return grid_search