Source code for catalyst.utils.torch

from typing import Dict, Iterable, List, Union  # isort:skip
import collections
import os
import re

import numpy as np
import safitty

import torch
from torch import nn
import torch.backends.cudnn as cudnn

from catalyst import utils
from catalyst.utils.typing import Device, Model, Optimizer


[docs]def ce_with_logits(logits, target): """Returns cross entropy for giving logits""" return torch.sum(-target * torch.log_softmax(logits, -1), -1)
[docs]def log1p_exp(x): """ Computationally stable function for computing log(1+exp(x)). """ x_ = x * x.ge(0).to(torch.float32) res = x_ + torch.log1p(torch.exp(-torch.abs(x))) return res
[docs]def normal_sample(mu, sigma): """ Sample from multivariate Gaussian distribution z ~ N(z|mu,sigma) while supporting backpropagation through its mean and variance. """ return mu + sigma * torch.randn_like(sigma)
[docs]def normal_logprob(mu, sigma, z): """ Probability density function of multivariate Gaussian distribution N(z|mu,sigma). """ normalization_constant = (-sigma.log() - 0.5 * np.log(2 * np.pi)) square_term = -0.5 * ((z - mu) / sigma)**2 logprob_vec = normalization_constant + square_term logprob = logprob_vec.sum(1) return logprob
[docs]def soft_update(target, source, tau): """Updates the target data with smoothing by ``tau``""" for target_param, param in zip(target.parameters(), source.parameters()): target_param.data.copy_( target_param.data * (1.0 - tau) + param.data * tau )
[docs]def get_optimizable_params(model_or_params): """ Returns all the parameters that requires gradients """ params: Iterable[torch.Tensor] = model_or_params if isinstance(model_or_params, nn.Module): params = model_or_params.parameters() master_params = [p for p in params if p.requires_grad] return master_params
[docs]def get_optimizer_momentum(optimizer: Optimizer) -> float: """ Get momentum of current optimizer. Args: optimizer: PyTorch optimizer Returns: float: momentum at first param group """ beta = safitty.get(optimizer.param_groups, 0, "betas", 0) momentum = safitty.get(optimizer.param_groups, 0, "momentum") return beta if beta is not None else momentum
[docs]def set_optimizer_momentum(optimizer: Optimizer, value: float, index: int = 0): """ Set momentum of ``index`` 'th param group of optimizer to ``value`` Args: optimizer: PyTorch optimizer value (float): new value of momentum index (int, optional): integer index of optimizer's param groups, default is 0 """ betas = safitty.get(optimizer.param_groups, index, "betas") momentum = safitty.get(optimizer.param_groups, index, "momentum") if betas is not None: _, beta = betas safitty.set( optimizer.param_groups, index, "betas", value=(value, beta) ) elif momentum is not None: safitty.set(optimizer.param_groups, index, "momentum", value=value)
[docs]def assert_fp16_available() -> None: """ Asserts for installed and available Apex FP16 """ assert torch.backends.cudnn.enabled, \ "fp16 mode requires cudnn backend to be enabled." try: import apex # noqa: F401 from apex import amp # noqa: F401 except ImportError: assert False, \ "NVidia Apex package must be installed. " \ "See https://github.com/NVIDIA/apex."
[docs]def get_device() -> torch.device: """ Simple returning the best available device (GPU or CPU) """ return torch.device("cuda" if torch.cuda.is_available() else "cpu")
[docs]def get_available_gpus(): """ Array of available GPU ids Returns: iterable: available GPU ids Examples: >>> os.environ["CUDA_VISIBLE_DEVICES"] = "0,2" >>> get_available_gpus() >>> [0, 2] >>> os.environ["CUDA_VISIBLE_DEVICES"] = "0,-1,1" >>> get_available_gpus() >>> [0] >>> os.environ["CUDA_VISIBLE_DEVICES"] = "" >>> get_available_gpus() >>> [] >>> os.environ["CUDA_VISIBLE_DEVICES"] = "-1" >>> get_available_gpus() >>> [] """ if "CUDA_VISIBLE_DEVICES" in os.environ: result = os.environ["CUDA_VISIBLE_DEVICES"].split(",") result = [int(id_) for id_ in result if id_ != ""] # invisible GPUs # https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#env-vars if -1 in result: index = result.index(-1) result = result[:index] elif torch.cuda.is_available(): result = list(range(torch.cuda.device_count())) else: result = [] return result
[docs]def get_activation_fn(activation: str = None): """ Returns the activation function from ``torch.nn`` by its name """ if activation is None or activation.lower() == "none": activation_fn = lambda x: x # noqa: E731 else: activation_fn = torch.nn.__dict__[activation]() return activation_fn
[docs]def any2device(value, device: Device): """ Move tensor, list of tensors, list of list of tensors, dict of tensors, tuple of tensors to target device. Args: value: Object to be moved device (Device): target device ids Returns: Same structure as value, but all tensors and np.arrays moved to device """ if isinstance(value, dict): return dict((k, any2device(v, device)) for k, v in value.items()) elif isinstance(value, (tuple, list)): return list(any2device(v, device) for v in value) elif torch.is_tensor(value): return value.to(device, non_blocking=True) elif isinstance(value, (np.ndarray, np.void)) \ and value.dtype.fields is not None: return dict( (k, any2device(value[k], device)) for k in value.dtype.fields.keys() ) elif isinstance(value, np.ndarray): return torch.Tensor(value).to(device) return value
[docs]def prepare_cudnn(deterministic: bool = None, benchmark: bool = None) -> None: """ Prepares CuDNN benchmark and sets CuDNN to be deterministic/non-deterministic mode Args: deterministic (bool): deterministic mode if running in CuDNN backend. benchmark (bool): If ``True`` use CuDNN heuristics to figure out which algorithm will be most performant for your model architecture and input. Setting it to ``False`` may slow down your training. """ if torch.cuda.is_available(): # CuDNN reproducibility # https://pytorch.org/docs/stable/notes/randomness.html#cudnn if deterministic is None: deterministic = \ os.environ.get("CUDNN_DETERMINISTIC", "True") == "True" cudnn.deterministic = deterministic # https://discuss.pytorch.org/t/how-should-i-disable-using-cudnn-in-my-code/38053/4 if benchmark is None: benchmark = os.environ.get("CUDNN_BENCHMARK", "True") == "True" cudnn.benchmark = benchmark
[docs]def process_model_params( model: Model, layerwise_params: Dict[str, dict] = None, no_bias_weight_decay: bool = True, lr_scaling: float = 1.0 ) -> List[Union[torch.nn.Parameter, dict]]: """ Gains model parameters for ``torch.optim.Optimizer`` Args: model (torch.nn.Module): Model to process layerwise_params (Dict): Order-sensitive dict where each key is regex pattern and values are layer-wise options for layers matching with a pattern no_bias_weight_decay (bool): If true, removes weight_decay for all ``bias`` parameters in the model lr_scaling (float): layer-wise learning rate scaling, if 1.0, learning rates will not be scaled Returns: iterable: parameters for an optimizer Examples: >>> model = catalyst.contrib.models.segmentation.ResnetUnet() >>> layerwise_params = collections.OrderedDict([ >>> ("conv1.*", dict(lr=0.001, weight_decay=0.0003)), >>> ("conv.*", dict(lr=0.002)) >>> ]) >>> params = process_model_params(model, layerwise_params) >>> optimizer = torch.optim.Adam(params, lr=0.0003) """ params = list(model.named_parameters()) layerwise_params = layerwise_params or collections.OrderedDict() model_params = [] for name, parameters in params: options = {} for pattern, options_ in layerwise_params.items(): if re.match(pattern, name) is not None: # all new LR rules write on top of the old ones options = utils.merge_dicts(options, options_) # no bias decay from https://arxiv.org/abs/1812.01187 if no_bias_weight_decay and name.endswith("bias"): options["weight_decay"] = 0.0 # lr linear scaling from https://arxiv.org/pdf/1706.02677.pdf if "lr" in options: options["lr"] *= lr_scaling model_params.append({"params": parameters, **options}) return model_params
[docs]def set_requires_grad(model: Model, requires_grad: bool): """ Sets the ``requires_grad`` value for all model parameters. Args: model (torch.nn.Module): Model requires_grad (bool): value Examples: >>> model = SimpleModel() >>> set_requires_grad(model, requires_grad=True) """ requires_grad = bool(requires_grad) for param in model.parameters(): param.requires_grad = requires_grad
[docs]def get_network_output(net: Model, *input_shapes): """ For each input shape returns an output tensor Args: net (Model): the model *args: variable length argument list of shapes """ inputs = [] for input_shape in input_shapes: if isinstance(input_shape, dict): input_t = {} for key, input_shape_ in input_shape.items(): input_t[key] = torch.Tensor(torch.randn((1, ) + input_shape_)) else: input_t = torch.Tensor(torch.randn((1, ) + input_shape)) inputs.append(input_t) output_t = net(*inputs) return output_t
[docs]def detach(tensor: torch.Tensor) -> np.ndarray: """ Detaches the input tensor to a numpy array """ return tensor.detach().cpu().numpy()
__all__ = [ "ce_with_logits", "log1p_exp", "normal_sample", "normal_logprob", "soft_update", "get_optimizable_params", "get_optimizer_momentum", "set_optimizer_momentum", "assert_fp16_available", "get_device", "get_available_gpus", "get_activation_fn", "any2device", "prepare_cudnn", "process_model_params", "set_requires_grad", "get_network_output", "detach" ]