def render(self, is_lock):
bimpy.indent(10)
bimpy.text('- Plane')
bimpy.same_line()
bimpy_tools.help_marker(
'generate with random points\n' \
'* plane random: random on whole plane\n' \
'* balanced random: balanced positive and negative samples'
)
bimpy.push_item_width(140)
if bimpy.begin_combo('strategy##plane_random_generator',
self._select_strategy):
for item in self._strategy_list:
is_selected = bimpy.Bool(self._select_strategy == item)
if bimpy.selectable(item, is_selected) and not is_lock:
self._select_strategy = item
if is_selected.value:
bimpy.set_item_default_focus()
bimpy.end_combo()
bimpy.pop_item_width()
bimpy.unindent(10)
def render(self, is_lock):
bimpy.set_next_tree_node_open(True, bimpy.Condition.FirstUseEver)
if not bimpy.tree_node('convex points##convex_component'):
return
bimpy.same_line()
bimpy_tools.help_marker('Convex points should be presented in counter-clockwise order')
flags = bimpy.InputTextFlags.EnterReturnsTrue
if is_lock:
flags |= bimpy.InputTextFlags.ReadOnly
last_convex_number_value = self._convex_number.value
if bimpy.input_int('number##convex_component', self._convex_number, 1, 1, flags):
self._convex_number.value = max(3, self._convex_number.value)
if last_convex_number_value > self._convex_number.value:
self._convex_data = self._convex_data[:self._convex_number.value] # cut back points
else:
self._convex_data.extend([
[bimpy.Float(0), bimpy.Float(0)]
for _ in range(last_convex_number_value, self._convex_number.value)
])
# show convex value setting
bimpy.set_next_tree_node_open(self._convex_number.value < 10, bimpy.Condition.FirstUseEver)
if bimpy.tree_node('convex value ({})##convex_component'.format(self._convex_number.value)):
for index in range(self._convex_number.value):
bimpy.push_item_width(210)
bimpy.input_float2(
'{:<3d}'.format(index),
self._convex_data[index][0],
self._convex_data[index][1],
flags=flags
)
bimpy.pop_item_width()
bimpy.tree_pop()
# draw part
bimpy.new_line()
if bimpy.button('draw convex##convex_component') and not is_lock:
self._convex_data_backup = [[item[0].value, item[1].value]
for item in self._convex_data]
self._convex_draw_flag = True
self._convex_data = []
self._convex_number.value = 0
bimpy.tree_pop()
def render(self, is_lock):
bimpy.indent(10)
bimpy.text('- SGD')
bimpy.same_line()
bimpy_tools.help_marker('torch.optim.SGD')
flags = bimpy.InputTextFlags.EnterReturnsTrue
if is_lock:
flags |= bimpy.InputTextFlags.ReadOnly
bimpy.push_item_width(120)
if bimpy.input_float('lr##sgd_optimizer', self._lr, flags=flags):
self._lr.value = max(0.0, self._lr.value)
if bimpy.input_float('momentum##sgd_optimizer',
self._momentum,
flags=flags):
self._momentum.value = max(0.0, self._momentum.value)
if bimpy.input_float('dampening##sgd_optimizer',
self._dampening,
flags=flags):
self._dampening.value = max(0.0, self._dampening.value)
if bimpy.input_float('weight_decay##sgd_optimizer',
self._weight_decay,
flags=flags):
self._weight_decay.value = max(0.0, self._weight_decay.value)
if bimpy.checkbox('nesterov##sgd_optimizer', self._nesterov):
self._hint_nesterov = False
if self._nesterov.value:
if self._momentum.value == 0 or self._dampening.value > 0:
self._nesterov.value = False
self._hint_nesterov = True
bimpy.same_line()
bimpy_tools.help_marker(
'Nesterov momentum requires a momentum and zero dampening',
self._hint_nesterov)
bimpy.pop_item_width()
bimpy.unindent(10)
def render(self, is_lock):
bimpy.indent(10)
bimpy.text('- Linear layer init')
bimpy.same_line()
bimpy_tools.help_marker(
'Initializer used in torch.nn.Linear, use Kaiming uniform')
bimpy.push_item_width(120)
flags = bimpy.InputTextFlags.EnterReturnsTrue
if is_lock:
flags |= bimpy.InputTextFlags.ReadOnly
if bimpy.input_float('a##sgd_optimizer', self._a, flags=flags):
self._a.value = max(0.0, self._a.value)
if bimpy.begin_combo('mode##linear_layer_init', self._select_mode):
for item in self._mode_list:
is_selected = bimpy.Bool(self._select_mode == item)
if bimpy.selectable(item, is_selected) and not is_lock:
self._select_mode = item
if is_selected.value:
bimpy.set_item_default_focus()
bimpy.end_combo()
if bimpy.begin_combo('nonlinearity##linear_layer_init',
self._select_nonlinearity):
for item in self._nonlinearity_list:
is_selected = bimpy.Bool(self._select_nonlinearity == item)
if bimpy.selectable(item, is_selected) and not is_lock:
self._select_nonlinearity = item
if is_selected.value:
bimpy.set_item_default_focus()
bimpy.end_combo()
bimpy.pop_item_width()
bimpy.unindent(10)
def render(self, is_lock):
bimpy.indent(10)
bimpy.text('- MSE loss')
bimpy.same_line()
bimpy_tools.help_marker('torch.nn.MSELoss')
bimpy.push_item_width(120)
if bimpy.begin_combo('reduction##mse_loss_fn', self._select_redution):
for item in self._reduction_list:
is_selected = bimpy.Bool(self._select_redution == item)
if bimpy.selectable(item, is_selected) and not is_lock:
self._select_redution = item
if is_selected.value:
bimpy.set_item_default_focus()
bimpy.end_combo()
bimpy.pop_item_width()
bimpy.unindent(10)
def render(self, is_lock):
bimpy.indent(10)
bimpy.text('- Adam')
bimpy.same_line()
bimpy_tools.help_marker('torch.optim.Adam')
flags = bimpy.InputTextFlags.EnterReturnsTrue
if is_lock:
flags |= bimpy.InputTextFlags.ReadOnly
bimpy.push_item_width(140)
if bimpy.input_float('lr##adam_optimizer', self._lr, flags=flags):
self._lr.value = max(0.0, self._lr.value)
if bimpy.input_float2('momentum##adam_optimizer',
self._betas_first,
self._betas_second,
flags=flags):
self._betas_first.value = max(0.0, self._betas_first.value)
self._betas_second.value = max(0.0, self._betas_second.value)
if bimpy.input_float('eps##adam_optimizer',
self._eps,
decimal_precision=8,
flags=flags):
self._dampening.value = max(0.0, self._eps.value)
if bimpy.input_float('weight_decay##adam_optimizer',
self._weight_decay,
flags=flags):
self._weight_decay.value = max(0.0, self._weight_decay.value)
bimpy.checkbox('amsgrad##adam_optimizer', self._amsgrad)
bimpy.pop_item_width()
bimpy.unindent(10)
def render(self, is_lock):
bimpy.indent(10)
bimpy.text('- Raw Point')
bimpy.same_line()
bimpy_tools.help_marker('generate with raw points')
bimpy.push_item_width(120)
if bimpy.begin_combo('strategy##raw_point_generator',
self._select_strategy):
for item in self._strategy_list:
is_selected = bimpy.Bool(self._select_strategy == item)
if bimpy.selectable(item, is_selected) and not is_lock:
self._select_strategy = item
if is_selected.value:
bimpy.set_item_default_focus()
bimpy.end_combo()
bimpy.pop_item_width()
bimpy.unindent(10)