def test_equation_discoverer_hyperopt():
B = -2.56; a = 0.4; ts = np.linspace(0.45, 0.87, 5)
ys = (ts+B)*np.exp(a*ts); xs = np.exp(a*ts)
data = np.hstack((ts.reshape(-1, 1), xs.reshape(-1, 1), ys.reshape(-1, 1)))
np.random.seed(20)
ED = EqDisco(data = data,
task = None,
task_type = "differential",
time_index = 0,
target_variable_index = -1,
variable_names=["t", "x", "y"],
sample_size = 2,
verbosity = 1,
estimation_settings={
# "optimizer": hyperopt_fit,
"optimizer": 'hyperopt',
"hyperopt_space_fn": hp.qnormal,
"hyperopt_space_args": (0.4, 0.5, 1/1000),
"hyperopt_max_evals": 100,
}
)
ED.generate_models()
ED.fit_models()
def assert_line(models, i, expr, error, tol=1e-9, n=100):
assert str(ED.models[i].get_full_expr())[:n] == expr[:n]
assert abs(ED.models[i].get_error() - error) < tol
assert_line(ED.models, 0, "y", 0.06518775248116751)
assert_line(ED.models, 1, "0.401*x + y", 2.5584839808071686e-07, n=6)
return
def test_equation_discoverer():
np.random.seed(0)
def f(x):
return 2.0 * (x[:,0] + 0.3)
X = np.linspace(-1, 1, 20).reshape(-1,1)
Y = f(X).reshape(-1,1)
data = np.hstack((X,Y))
ED = EqDisco(data = data,
task = None,
target_variable_index = -1,
sample_size = 2,
verbosity = 1)
ED.generate_models()
ED.fit_models()
#print(ED.models[0].get_error())
assert np.abs(ED.models[0].get_error() - 0.72842105) < 1e-6
assert np.abs(ED.models[1].get_error() - 0.59163899) < 1e-6
def test_equation_discoverer_ODE():
B = -2.56; a = 0.4; ts = np.linspace(0.45, 0.87, 5)
ys = (ts+B)*np.exp(a*ts); xs = np.exp(a*ts)
data = np.hstack((ts.reshape(-1, 1), xs.reshape(-1, 1), ys.reshape(-1, 1)))
np.random.seed(20)
ED = EqDisco(data = data,
task = None,
task_type = "differential",
time_index = 0,
target_variable_index = -1,
variable_names=["t", "x", "y"],
sample_size = 2,
verbosity = 1)
ED.generate_models()
ED.fit_models()
def assert_line(models, i, expr, error, tol=1e-9, n=100):
assert str(models[i].get_full_expr())[:n] == expr[:n]
assert abs(models[i].get_error() - error) < tol
assert_line(ED.models, 0, "y", 0.06518775248116751)
assert_line(ED.models, 1, "0.400266188520229*x + y", 2.5265653001321905e-09, n=6)
return
from ProGED.equation_discoverer import EqDisco
from ProGED.parameter_estimation import DE_fit, hyperopt_fit #, DE_fit_metamodel
np.random.seed(0)
ED = EqDisco(data = data,
task = None,
task_type = "differential",
time_index = 0,
target_variable_index = aquation[0][0], # aquation = [123] -> target = 1 -> ([t,x,y,z]->x)
variable_names=["t", "x", "y", "z"],
generator = "grammar",
generator_template_name = "polynomial",
generator_settings={
# "variables": ["'x'", "'y'"],
"p_S": [0.4, 0.6],
"p_T": [0.4, 0.6],
"p_vars": [0.33, 0.33, 0.34],
"p_R": [1, 0],
"p_F": [],
"functions": [],
},
sample_size = sample_size,
verbosity = 4)
ED.generate_models()
ED.fit_models(
estimation_settings={
"timeout": 115,
"max_ode_steps": 10**6,
def oeis_eq_disco(seq_id: str, is_direct: bool, order: int):
"""Run eq. discovery of given OEIS sequence."""
data = grid(order, np.array(list(csv[seq_id])), is_direct)
variable_names_ = [f"an_{i}" for i in range(order, 0, -1)] + ["an"]
variable_names = ["n"] + variable_names_ if is_direct else variable_names_
variables = [f"'{an_i}'" for an_i in variable_names[:-1]]
# print(variable_names)
# print(variables)
# print(data.shape, type(data), data)
np.random.seed(random_seed)
ED = EqDisco(
data=data,
task=None,
target_variable_index=-1,
# variable_names=["an_2", "an_1", "an"],
variable_names=variable_names,
# sample_size = 16, # for recursive
# sample_size = 10, # for direct fib
sample_size=sample_size,
# sample_size = 50, # for recursive
# sample_size = 38, # for recursive
# sample_size = 100, # for recursive
verbosity=0,
# verbosity = 3,
generator="grammar",
generator_template_name=grammar_template_name,
# generator_settings={"variables": ["'an_2'", "'an_1'"],
generator_settings={
"variables": variables,
"p_T": p_T,
"p_R": p_R
},
estimation_settings={
# "verbosity": 3,
"verbosity": 1,
# "verbosity": 0,
"task_type": task_type,
# "task_type": "algebraic",
# "task_type": "integer-algebraic",
# "task_type": "oeis_recursive_error", # bad idea
"lower_upper_bounds": lower_upper_bounds,
# "lower_upper_bounds": (-1000, 1000), # najde (1001 pa ne) rec fib
# "lower_upper_bounds": (-100, 100), # ne najde DE
# "lower_upper_bounds": (-25, 25), # DA ne najde
# "lower_upper_bounds": (-11, 11), # shgo limit
# "lower_upper_bounds": (-14, 14), # DA dela
# "lower_upper_bounds": (-2, 2), # for direct
# "lower_upper_bounds": (-4, 4), # for direct fib
# "lower_upper_bounds": (-5, 5),
# "lower_upper_bounds": (-8, 8), # long enough for brute
# "optimizer": 'differential_evolution',
"optimizer": optimizer,
# "optimizer": 'hyperopt',
"timeout": timeout,
# "timeout": 1,
# "timeout": 13,
# "timeout_privilege": 30,
## hyperopt:
# "hyperopt_max_evals": 3250,
# "hyperopt_max_evals": 550, # finds if result=min(10**6, hyperopt...)
# "hyperopt_max_evals": 50,
# "hyperopt_max_evals": 750,
# "hyperopt_max_evals": 2000,
# "hyperopt_max_evals": 700,
# "hyperopt_max_evals": 500,
# "hyperopt_max_evals": 300,
# "hyperopt_space_fn": hp.randint,
# "hyperopt_space_fn": hp.choice,
# "hyperopt_space_fn": hp.loguniform, # Seems working, but loss 785963174750.8921 in 2000 evals.
# "hyperopt_space_fn": hp.qnormal,
# "hyperopt_space_fn": hp.normal,
# "hyperopt_space_args": (lower_upper_bounds[0], lower_upper_bounds[1]),
# "hyperopt_space_args": ([-2, -1, 0, 1],),
# "hyperopt_space_args": (0, 5, 1),
# "hyperopt_space_args": (0, 2), # Seems working for hp.normal, but loss 1492702199922.7131 in 2000 evals.
# "hyperopt_space_args": [lower_upper_bounds[0]],
# "hyperopt_space_kwargs": {"high": lower_upper_bounds[1]},
# "hyperopt_algo": tpe.suggest, # Works, with 1000 evals and (-2, 2) finds recursive.
# "hyperopt_algo": rand.suggest,
})
# for i in range(0, 100):
# np.random.seed(i)
# print("seed", i)
# print(ED)
# ED.generate_models()
# print(ED.models)
# ED.models = None
# 1/0
ED.generate_models()
print(ED.models)
# 1/0
seq_start = time.perf_counter()
ED.fit_models()
seq_cpu_time = time.perf_counter() - seq_start
print(f"\nParameter fitting for sequence {seq_id} "
f"took {seq_cpu_time} secconds.")
print("\nFinal score:")
for m in ED.models:
print(
f"model: {str(m.get_full_expr()):<30}; error: {m.get_error():<15}")
return
# 131,137,139,149,151,157,163,167,173,179,181,191,
# 193,197,199,211,223,227,229,233,239,241,251,257,
# 263,269,271][:40+1]
# oeis = oeis_primes
fibs = np.array(oeis).reshape(-1, 1)
ts = np.array([i for i in range(40+1)]).reshape(-1, 1)
data = np.hstack((ts, fibs))
np.random.seed(0)
ED = EqDisco(data = data,
task = None,
target_variable_index = -1,
variable_names=["n", "an"],
sample_size = 10,
verbosity = 0,
generator = "grammar",
generator_template_name = "polynomial",
generator_settings={"variables":["'n'"]},
estimation_settings={"verbosity": 0, "task_type": "algebraic", "lower_upper_bounds":
# (-5, 5)}
# (-10,8)} # Last bound where it still finds.
(0,1)} # Returns wrong error on windows.
)
ED.generate_models()
ED.fit_models()
# print(ED.models)
print("\n\nFinal score:")
for m in ED.models:
print(f"model: {str(m.get_full_expr()):<30}; error: {m.get_error():<15}")
phi = (1+5**(1/2))/2
psi = (1-5**(1/2))/2
p_R = [0.9, 0.1]
# # grammar = grammar_from_template("polynomial", {"variables": variables, "p_R": p_R, "p_T": p_T})
# # np.random.seed(0)
# # # print(grammar.generate_one())
# # # models = generate_models(grammar, symbols, strategy_settings = {"N":500})
# # # print(models)
np.random.seed(0)
# seed 0 , size 20 (16)
# seed3 size 15 an-1 + an-2 + c3
ED = EqDisco(data = data,
task = None,
target_variable_index = -1,
variable_names=["an_2", "an_1", "an"],
sample_size = 16,
verbosity = 0,
generator = "grammar",
generator_template_name = "polynomial",
generator_settings={"variables": ["'an_2'", "'an_1'"], "p_T": p_T, "p_R": p_R}
,estimation_settings={"verbosity": 1, "task_type": "algebraic", "lower_upper_bounds":
(0, 2)} # meja, ko se najde priblizno: (-10,8)}# , "timeout": np.inf}
)
# # print(data, data.shape)
ED.generate_models()
ED.fit_models()
# # try:
# # print(12/0)
print(ED.models)
# # print(ED.get_results())
# # print(ED.get_stats())
# print("\nFinal score:")