def run(dataset_train, dataset_test, pop_size, gens, cross_rate, fb, max_time=1200):
Xtrain, ytrain = dataset_train[:, :-1], dataset_train[:, -1]
Xtest, ytest = dataset_test[:, :-1], dataset_test[:, -1]
est_gp = Feat(obj="fitness,complexity",
pop_size=pop_size,
gens=gens,
max_time=max_time,
max_stall=50,
batch_size=10000,
ml = "LinearRidgeRegression",
sel='lexicase',
surv='nsga2',
max_depth=10,
max_dim=min([Xtrain.shape[1]*2,50]),
#random_state=random_seed,
functions="+,-,*,/,sqrt,sin,cos,tanh,exp,log,^,x,kd",
otype="f",
backprop=True,
iters=10,
n_threads=1,
verbosity=1,
# tuned parameters
cross_rate= cross_rate,
fb = fb,
root_xo_rate = 0.75,
softmax_norm = False
)
est_gp.fit(Xtrain, ytrain)
return RMSE(est_gp.predict(Xtrain), ytrain), RMSE(est_gp.predict(Xtest), ytest), est_gp.get_model()
def test_saving_loading(self):
self.debug("Pickle Feat object")
reg = clone(self.reg)
reg.fit(self.X, self.yr)
initial_pred = reg.predict(self.X)
reg.save('Feat_tmp.json')
loaded_reg = Feat().load('Feat_tmp.json')
# print('loaded_reg:',type(loaded_reg).__name__)
loaded_pred = loaded_reg.predict(self.X)
# print('initial pred:',initial_pred)
# print('loaded pred:',loaded_pred)
diff = np.abs(initial_pred-loaded_pred)
for i,d in enumerate(diff):
if d > 0.0001:
print('pred:',initial_pred[i],'loaded:',loaded_pred[i],
'diff:',d)
assert(d < 0.0001)
# assert(all([ip==lp for ip,lp in zip(initial_pred, loaded_pred)]))
assert(reg.get_representation() == loaded_reg.get_representation())
assert(reg.get_model() == loaded_reg.get_model())
assert((reg.get_coefs() == loaded_reg.get_coefs()).all())
loaded_params = loaded_reg.get_params()
# print('\n',10*'=','\n')
# print('loaded_params:')
# for k,v in loaded_params.items():
# print(k,':',v)
for k,v in reg.get_params().items():
if k not in loaded_params.keys():
print(k,'not in ',loaded_params.keys())
assert(k in loaded_params.keys())
if isinstance(v,float):
if np.abs(loaded_params[k] - v) > 0.0001:
print('loaded_params[',k,'] =',
loaded_params[k], '\nwhich is different from:', v)
assert(np.abs(loaded_params[k] - v) < 0.0001)
elif loaded_params[k] != v:
print('loaded_params[',k,'] =',
loaded_params[k], '\nwhich is different from:', v)
assert(loaded_params[k] == v)
loaded_reg.fit(self.X, self.yr)
for train_idx, test_idx in kf.split(X, y):
# print('train_idx:',train_idx)
clf.fit(X[train_idx], y[train_idx], zfile, train_idx)
scores.append(clf.score(X[test_idx], y[test_idx], zfile, test_idx))
print('scores:', scores)
###################################################################################################
# fit to all data
###################################################################################################
print('fitting longer to all data...')
clf.gens = 20
clf.verbosity = 2
clf.fit(X, y, zfile, np.arange(len(X)))
print('model:', clf.get_model())
##################################################################################################
# Plot t-SNE transformation
###################################################################################################
print('transform:')
Phi = clf.transform(X, zfile, np.arange(len(X)))
# use t-SNE to visualize transformation
import sklearn
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt
import seaborn as sns
import matplotlib.patheffects as PathEffects
proj = TSNE(random_state=42).fit_transform(Phi)