def grad_shower(func: ExtendedFunction,
start_point: np.ndarray,
strategy_sup: Callable[[ExtendedFunction], StepStrategy],
max_iters=15,
initial_steps=(0.01, 0.1, 1, 10)
):
strategy = strategy_sup(func)
data = [draw_grad_des(func, start_point, strategy, initial_step=s, max_iters=max_iters) for s in initial_steps]
draw_grid(data, ncols=2)
def draw_test_grad():
data = [draw_one_test_grad(n) for n in range(2, 6)]
draw_grid(data, ncols=2)
results_yy.append(res.x)
iter_by_eps[e_name] = (xx, iter_yy)
calls_by_eps[e_name] = (xx, calls_yy)
results_by_eps[e_name] = (xx, results_yy)
grid = [
by_eps_plot("Iterations by eps", iter_by_eps),
by_eps_plot("Calls by eps", calls_by_eps),
by_eps_plot("Result by eps", results_by_eps)
]
for e_name in ENGINES.keys():
grid.append(range_plt(e_name, 1e-4))
draw_grid(grid, ncols=engines_cnt)
# %%
"""
### Покажем результат работы метода градиентного спуска на некоторых примерах
"""
# %%
def draw_grad_des(e_func: ExtendedFunction,
start: np.ndarray,
strategy: StepStrategy,
max_iters=30,
initial_step=1):
def _drawer(ax):
grad = GradDescent()