def model_summarize(model, show_weights=True, show_parameters=True):
"""Summarizes torch model by showing trainable parameters and weights."""
tmpstr = model.__class__.__name__ + ' (\n'
params_num = 0
for key, module in model._modules.items():
# if it contains layers let call it recursively to get params and weights
if type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential
]:
modstr = model_summarize(module)
else:
modstr = module.__repr__()
if isinstance(modstr, str):
modstr = _addindent(modstr, 2)
elif isinstance(modstr, tuple):
modstr = _addindent(modstr[0], 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
weights = tuple([tuple(p.size()) for p in module.parameters()])
# rest = b if b > 0 else params
# params_num = params_num + rest
params_num += params
tmpstr += ' (' + key + '): ' + modstr
if show_weights:
tmpstr += ', weights={}'.format(weights)
if show_parameters:
tmpstr += ', parameters={}'.format(params)
tmpstr += '\n'
tmpstr = tmpstr + ')'
return tmpstr, params_num * 4. / (1024**2)
def __repr__(self):
main_str = self._get_name() + '('
if self.regular is not None:
mod_str = _addindent(repr(self.regular), 2)
main_str += '\n (Regular & Fusion): ' + mod_str
if self.fuse is not None:
mod_str = _addindent(repr(self.fuse), 2)
main_str += '\n (Fuse): ' + mod_str
if self.combined is not None:
mod_str = _addindent(repr(self.combined), 2)
main_str += '\n (Combined): ' + mod_str
return main_str
def torch_summarize(model, show_weights=True, show_parameters=True):
"""
SOURCE: https://stackoverflow.com/a/45528544
Summarizes torch model by showing trainable parameters and weights.
"""
tmpstr = model.__class__.__name__ + ' (\n'
for key, module in model._modules.items():
# if it contains layers let call it recursively to get params and weights
if type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential
]:
modstr = torch_summarize(module)
else:
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += ' (' + key + '): ' + modstr
if show_weights:
tmpstr += ', weights={}'.format(weights)
if show_parameters:
tmpstr += ', parameters={}'.format(params)
tmpstr += '\n'
tmpstr = tmpstr + ')'
print("\n\n\n" + 20 * "=")
print("Total number of params: {}\n".format(
sum([param.nelement() for param in model.parameters()])))
print(tmpstr)
print(20 * "=" + "\n\n")
def __repr__(self):
# We don't want to show `LinearPackedParams` children, hence custom
# `__repr__`. This is the same as nn.Module.__repr__, except the check
# for the `LinearPackedParams`.
# You should still override `extra_repr` to add more info.
extra_lines = []
extra_repr = self.extra_repr()
# empty string will be split into list ['']
if extra_repr:
extra_lines = extra_repr.split('\n')
child_lines = []
for key, module in self._modules.items():
mod_str = repr(module)
mod_str = _addindent(mod_str, 2)
child_lines.append('(' + key + '): ' + mod_str)
lines = extra_lines + child_lines
main_str = self._get_name() + '('
if lines:
# simple one-liner info, which most builtin Modules will use
if len(extra_lines) == 1 and not child_lines:
main_str += extra_lines[0]
else:
main_str += '\n ' + '\n '.join(lines) + '\n'
main_str += ')'
return main_str
def summarize(model, num_params, show_weights=True, show_parameters=True):
tmpstr = model.__class__.__name__ + ' (\n'
for key, module in model._modules.items():
# if it contains layers let call it recursively to get params and weights
if type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential
]:
modstr, num_params_rec = summarize(module, num_params,
show_weights,
show_parameters)
num_params += num_params_rec
else:
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
num_params += params
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += ' (' + key + '): ' + modstr
if show_weights:
tmpstr += ', weights={}'.format(weights)
if show_parameters:
tmpstr += ', parameters={}'.format(params)
tmpstr += '\n'
tmpstr = tmpstr + ')'
return tmpstr, num_params
def torch_summarize(model, show_weights=True, show_parameters=True):
"""
This function will give an overview of the created neural network
by printing out all the weights and parameters
"""
tmpstr = model.__class__.__name__ + ' (\n'
for key, module in model._modules.items():
# if it contains layers call it recursively to get params and weights
if type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential
]:
modstr = torch_summarize(module)
else:
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += ' (' + key + '): ' + modstr
if show_weights:
tmpstr += ', weights={}'.format(weights)
if show_parameters:
tmpstr += ', parameters={}'.format(params)
tmpstr += '\n'
tmpstr = tmpstr + ')'
return tmpstr
def repr(model):
# We treat the extra repr like the sub-module, one item per line
extra_lines = []
extra_repr = model.extra_repr()
# empty string will be split into list ['']
if extra_repr:
extra_lines = extra_repr.split('\n')
child_lines = []
total_params = 0
for key, module in model._modules.items():
mod_str, num_params = repr(module)
mod_str = _addindent(mod_str, 2)
child_lines.append('(' + key + '): ' + mod_str)
total_params += num_params
lines = extra_lines + child_lines
for name, p in model._parameters.items():
if hasattr(p, 'shape'):
total_params += reduce(lambda x, y: x * y, p.shape)
main_str = model._get_name() + '('
if lines:
# simple one-liner info, which most builtin Modules will use
if len(extra_lines) == 1 and not child_lines:
main_str += extra_lines[0]
else:
main_str += '\n ' + '\n '.join(lines) + '\n'
main_str += ')'
if file is sys.stdout:
main_str += ', \033[92m{:,}\033[0m params'.format(total_params)
else:
main_str += ', {:,} params'.format(total_params)
return main_str, total_params
def summarize(cls, model, params_switch=True, weights_switch=True):
"""
Summarizes torch model by showing trainable parameters and weights.
"""
from torch.nn.modules.module import _addindent
tmpstr = model.__class__.__name__ + '(\n'
for key, module in model._modules.items():
if type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential
]:
modstr = cls.summarize(module)
else:
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
tmpstr += ' (' + key + '): ' + modstr
if params_switch:
params = sum([np.prod(p.size()) for p in module.parameters()])
tmpstr += ', parameters={}'.format(params)
if weights_switch:
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += ', weights={}'.format(weights)
tmpstr += '\n'
tmpstr = tmpstr + ')'
return tmpstr
def torch_summarize(model, show_weights=True, show_parameters=True):
"""Summarizes torch model by showing trainable parameters and weights."""
tmpstr = model.__class__.__name__ + ' (\n'
for key, module in model._modules.items():
# if it contains layers let call it recursively to get params and
# weights
if type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential
]:
modstr = torch_summarize(module)
else:
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += ' (' + key + '): ' + modstr
if show_weights:
tmpstr += ', weights={}'.format(weights)
if show_parameters:
tmpstr += ', parameters={}'.format(params)
tmpstr += '\n'
tmpstr = tmpstr + ')'
return tmpstr
def hide_packed_params_repr(self, params):
# We don't want to show `PackedParams` children, hence custom
# `__repr__`. This is the same as nn.Module.__repr__, except the check
# for the `params module`.
extra_lines = []
extra_repr = self.extra_repr()
# empty string will be split into list ['']
if extra_repr:
extra_lines = extra_repr.split('\n')
child_lines = []
for key, module in self._modules.items():
if isinstance(module, params):
continue
mod_str = repr(module)
mod_str = _addindent(mod_str, 2)
child_lines.append('(' + key + '): ' + mod_str)
lines = extra_lines + child_lines
main_str = self._get_name() + '('
if lines:
# simple one-liner info, which most builtin Modules will use
if len(extra_lines) == 1 and not child_lines:
main_str += extra_lines[0]
else:
main_str += '\n ' + '\n '.join(lines) + '\n'
main_str += ')'
return main_str
def torch_summarize(model, show_weights=False, show_parameters=True):
"""
Summarizes torch model by showing trainable parameters and weights
author: wassname
url: https://gist.github.com/wassname/0fb8f95e4272e6bdd27bd7df386716b7
license: MIT
"""
tmpstr = model.__class__.__name__ + ' (\n'
for key, module in model.network._modules.items():
# if it contains layers let call it recursively to get params and weights
is_container = type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential, torch.nn.Module
]
if is_container:
modstr = torch_summarize(module)
else:
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += ' (' + key + '): ' + modstr
if show_weights:
tmpstr += ', weights={}'.format(weights)
if show_parameters:
tmpstr += ', parameters={}'.format(params)
tmpstr += '\n'
if is_container:
tmpstr += '\n'
tmpstr = tmpstr + ')'
return tmpstr
def torch_summarize(self, model, show_weights=True, show_parameters=True):
"""Summarizes torch model by showing trainable parameters and weights.
Taken from wassname on Stackoverflow: https://stackoverflow.com/a/45528544
"""
tmpstr = model.__class__.__name__ + ' (\n'
for key, module in model._modules.items():
# if it contains layers let call it recursively to get params and weights
if type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential
]:
modstr = torch_summarize(module)
else:
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += ' (' + key + '): ' + modstr
if show_weights:
tmpstr += ', weights={}'.format(weights)
if show_parameters:
tmpstr += ', parameters={}'.format(params)
tmpstr += '\n'
tmpstr = tmpstr + ')'
return tmpstr
def __repr__(self):
# We treat the extra repr like the sub-module, one item per line
extra_lines = []
extra_repr = self.extra_repr()
# empty string will be split into list ['']
if extra_repr:
extra_lines = extra_repr.split('\n')
child_lines = []
for key, module in self._modules.items():
mod_str = repr(module)
mod_str = _addindent(mod_str, 2)
child_lines.append('(' + key + '): ' + mod_str)
lines = extra_lines + child_lines
main_str = self._get_name() + '\tnum_heads:' + str(
self.num_heads) + '\t qkv_dim:' + str(self.qkv_dim)
if lines:
# simple one-liner info, which most builtin Modules will use
if len(extra_lines) == 1 and not child_lines:
main_str += extra_lines[0]
else:
main_str += '\n ' + '\n '.join(lines) + '\n'
main_str += ')'
return main_str
def torch_summarize(model, show_weights=True, show_parameters=True):
"""Resume el modelo torch mostrando parámetros y pesos entrenables."""
tmpstr = model.__class__.__name__ + ' (\n'
for key, module in model._modules.items():
# Si contiene capas, llamemos recursivamente para obtener parámetros y pesos
if type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential
]:
modstr = torch_summarize(module)
else:
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += ' (' + key + '): ' + modstr
if show_weights:
tmpstr += ', pesos={}'.format(weights)
if show_parameters:
tmpstr += ', parametros={}'.format(params)
tmpstr += '\n'
tmpstr = tmpstr + ')'
return tmpstr
def summary(model, show_weights=True, show_parameters=True):
"""Summarizes torch model by showing trainable parameters and weights."""
tmpstr = model.__class__.__name__ + " (\n"
for key, module in model._modules.items():
# if it contains layers let call it recursively to get params and weights
if type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential,
]:
modstr = torch_summarize(module)
else:
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += " (" + key + "): " + modstr
if show_weights:
tmpstr += ", weights={}".format(weights)
if show_parameters:
tmpstr += ", parameters={}".format(params)
tmpstr += "\n"
tmpstr = tmpstr + ")"
return tmpstr
def print_summary(self, show_weights=False, show_parameters=False):
"""
Summarizes torch model by showing trainable parameters and weights.
"""
tmpstr = self.__class__.__name__ + ' (\n'
for key, module in self._modules.items():
# if it contains layers let call it recursively to get params and weights
if type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential
]:
modstr = self.print_summary()
else:
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += ' (' + key + '): ' + modstr
if show_weights:
tmpstr += ', weights={}'.format(weights)
if show_parameters:
tmpstr += ', parameters={}'.format(params)
tmpstr += '\n'
tmpstr = tmpstr + ')'
print(f'========== {self.name} Model Summary ==========')
print(tmpstr)
num_params = sum(p.numel() for p in self.parameters()
if p.requires_grad)
print(f"Params #: {'{:,}'.format(num_params)}")
print('==================================================')
def repr(model):
# We treat the extra repr like the sub-module, one item per line
extra_lines = []
extra_repr = model.extra_repr()
# empty string will be split into list ['']
if extra_repr:
extra_lines = extra_repr.split("\n")
child_lines = []
total_params = 0
for key, module in model._modules.items():
mod_str, num_params = repr(module)
mod_str = _addindent(mod_str, 2)
child_lines.append("(" + key + "): " + mod_str)
total_params += num_params
lines = extra_lines + child_lines
for name, p in model._parameters.items():
if p is not None:
total_params += reduce(lambda x, y: x * y, p.shape)
main_str = model._get_name() + "("
if lines:
# simple one-liner info, which most builtin Modules will use
if len(extra_lines) == 1 and not child_lines:
main_str += extra_lines[0]
else:
main_str += "\n " + "\n ".join(lines) + "\n"
main_str += ")"
main_str += ", {:,} params".format(total_params)
return main_str, total_params
def torch_summarize(self, show_weights=True, show_parameters=True):
"""Summarizes torch model by showing trainable parameters and weights."""
tmpstr = self.network.__class__.__name__ + ' (\n'
total_params = 0
total_params_wo_embedding = 0
for key, module in self.network._modules.items():
# if it contains layers let call it recursively to get params and weights
if type(module) in [
nn.modules.container.Container,
nn.modules.container.Sequential
]:
modstr = self.torch_summarize(self.network)
else:
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
total_params += params
if key != 'encoder':
total_params_wo_embedding += params
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += ' (' + key + '): ' + modstr
if show_weights:
tmpstr += ', weights={}'.format(weights)
if show_parameters:
tmpstr += ', parameters={}'.format(params)
tmpstr += '\n'
tmpstr += ')\ntotal_params: %d, total_params_wo_embedding: %d' % \
(total_params, total_params_wo_embedding)
self.num_free_params = total_params_wo_embedding
return tmpstr
def torch_summarize(model, show_weights=True, show_parameters=True):
"""Summarizes torch model by showing trainable parameters and weights."""
tmpstr = model.__class__.__name__ + ' (\n'
total_params = 0
for key, module in model._modules.items():
# if it contains layers let call it recursively to get params and weights
if type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential
]:
modstr = torch_summarize(module)
else:
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
total_params += params
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += ' (' + key + '): ' + modstr
if show_weights:
tmpstr += ', weights={}'.format(weights)
if show_parameters:
tmpstr += ', parameters={}'.format(params)
tmpstr += '\n'
tmpstr = tmpstr + ')'
tmpstr += '\n {} learnable parameters'.format(total_params)
return tmpstr
def torch_summarize(model, show_weights=True, show_parameters=True):
"""Summarizes torch model by showing trainable parameters and weights."""
tmpstr = model.__class__.__name__ + ' (\n'
parameters = 0
convs = 0
for key, module in model._modules.items():
# if it contains layers let call it recursively to get params and weights
if type(module) in [torch.nn.modules.container.Container, torch.nn.modules.container.Sequential]:
modstr, p, cnvs = torch_summarize(module)
parameters += p
convs += cnvs
else:
modstr = module.__repr__()
convs += len(modstr.split('Conv2d')) - 1
modstr = _addindent(modstr, 2)
# if 'conv' in key:
# convs += 1
params = sum([np.prod(p.size()) for p in module.parameters()])
parameters += params
weights = tuple([tuple(p.size()) for p in module.parameters()])
tmpstr += ' (' + key + '): ' + modstr
if show_weights:
tmpstr += ', weights={}'.format(weights)
if show_parameters:
tmpstr += ', parameters={} / {}'.format(params, parameters)
tmpstr += ', convs={}'.format(convs)
tmpstr += '\n'
tmpstr = tmpstr + ')'
return tmpstr, parameters, convs
def nested_str(self) -> str:
"""
Return the Python code generated for current GraphModule and its children GraphModules
"""
module_code = self.code
module_code = module_code.lstrip('\n')
module_code = f"class {self._get_name()}(torch.nn.Module):\n" + module_code
module_code = _addindent(module_code, 4)
submodule_code_list = [""]
for submodule in self.children():
if isinstance(submodule, GraphModule):
submodule_code_list.append(submodule.__nested_code())
submodule_code = "\n".join(submodule_code_list)
submodule_code = _addindent(submodule_code, 4)
return module_code + submodule_code
def torch_summarize(model,
show_weights=True,
show_parameters=True,
show_trainable=True):
"""Summarizes torch model by showing trainable parameters and weights."""
tmpstr = model.__class__.__name__ + " (\n"
for key, module in model._modules.items():
# If it contains layers let call it recursively to get params and weights
if type(module) in [
torch.nn.modules.container.Container,
torch.nn.modules.container.Sequential
]:
modstr = torch_summarize(module)
else:
modstr = module.__repr__()
# ====================== Extra stuff (for displaying nn.Parameter) ======================
lst_params = []
for name, p in module.named_parameters():
if (type(p) == torch.nn.parameter.Parameter
and "weight" not in name and "bias" not in name):
lst_params.append(" ({}): Parameter{}".format(
name, tuple(p.size())))
if (lst_params):
modstr = modstr[:-1]
modstr += "\n".join(lst_params)
modstr += "\n)"
# ====================== Extra stuff (for displaying nn.Parameter) ======================
modstr = _addindent(modstr, 2)
weights = tuple([tuple(p.size()) for p in module.parameters()])
params = sum([np.prod(p.size()) for p in module.parameters()])
total_params = sum([
torch.LongTensor(list(p.size())).prod()
for p in module.parameters()
])
trainable_params = sum([
torch.LongTensor(list(p.size())).prod()
for p in module.parameters() if p.requires_grad
])
tmpstr += " (" + key + "): " + modstr
if show_weights:
tmpstr += ", weights = {}".format(weights)
if show_parameters:
tmpstr += ", parameters = {:,}".format(params)
if show_trainable and total_params != 0 and total_params == trainable_params:
tmpstr += " (Trainable)"
tmpstr += "\n"
tmpstr = tmpstr + ")"
return tmpstr
def __repr__(self):
""" Overwrite the repr method to get a nice representation."""
tmpstr = self.__class__.__name__ + '(\n'
for key, module in self._modules.items():
modstr = module.__repr__()
modstr = _addindent(modstr, 2)
tmpstr = tmpstr + ' ({}, {}): '.format(
key, self._num_parameters(module)) + modstr + '\n'
tmpstr = tmpstr + ')'
return tmpstr
def __repr__(self):
# A simplified version of torch.nn.Module.__repr__
from torch.nn.modules.module import _addindent
child_lines = []
for idx, module in enumerate(self._modules):
mod_str = repr(module)
mod_str = _addindent(mod_str, 2)
child_lines.append('(' + str(idx) + '): ' + mod_str)
main_str = self.__class__.__name__ + '('
if child_lines:
main_str += '\n ' + '\n '.join(child_lines) + '\n'
main_str += ')'
return main_str
def get_model_parameter_str(model):
"""Get a summary string with model parameters
Adapted from https://stackoverflow.com/a/45528544
"""
def is_leaf(module):
if isinstance(module, (nn.ModuleList, nn.Sequential)):
return False
elif any((isinstance(module, (nn.ModuleList, nn.Sequential))
for module in module._modules.values())):
return False
return all((is_leaf(m) for m in module._modules.values()))
from torch.nn.modules.module import _addindent
tmpstr = model.__class__.__name__ + ' (\n'
total_params = 0
for key, module in model._modules.items():
if module is None:
continue
if is_leaf(module):
modstr = module.__repr__()
leaf = True
else:
modstr = get_model_parameter_str(module)
leaf = False
modstr = _addindent(modstr, 2)
params = sum([np.prod(p.size()) for p in module.parameters()])
total_params += params
tmpstr += ' (' + key + '): ' + modstr
if leaf and params > 0:
tmpstr += ', parameters={}'.format(params)
tmpstr += '\n'
tmpstr = tmpstr + ')'
if total_params > 0:
tmpstr += ', parameters={}'.format(total_params)
return tmpstr
