qualia2 package

Subpackages

• qualia2.data package
  • Submodules
  • qualia2.data.basic module
  • qualia2.data.cifar module
  • qualia2.data.dataloader module
  • qualia2.data.dataset module
  • qualia2.data.emnist module
  • qualia2.data.fashion_mnist module
  • qualia2.data.fimlp module
  • qualia2.data.kuzushi_mnist module
  • qualia2.data.mnist module
  • qualia2.data.stl10 module
  • qualia2.data.transforms module
  • Module contents
• qualia2.functions package
  • Submodules
  • qualia2.functions.activation module
  • qualia2.functions.array module
  • qualia2.functions.complex module
  • qualia2.functions.conv module
  • qualia2.functions.distance module
  • qualia2.functions.dropout module
  • qualia2.functions.linear_algebra module
  • qualia2.functions.loss module
  • qualia2.functions.normalize module
  • qualia2.functions.pool module
  • qualia2.functions.recurrent module
  • qualia2.functions.sparse module
  • qualia2.functions.trigonometric module
  • qualia2.functions.ufunc module
  • qualia2.functions.vision module
  • Module contents
• qualia2.nn package
  • Subpackages
    • qualia2.nn.modules package
      • Submodules
      • qualia2.nn.modules.activation module
      • qualia2.nn.modules.conv module
      • qualia2.nn.modules.dropout module
      • qualia2.nn.modules.linear module
      • qualia2.nn.modules.module module
      • qualia2.nn.modules.normalize module
      • qualia2.nn.modules.pool module
      • qualia2.nn.modules.recurrent module
      • qualia2.nn.modules.sparse module
      • Module contents
  • Submodules
  • qualia2.nn.init module
  • qualia2.nn.optim module
  • Module contents
• qualia2.rl package
  • Subpackages
    • qualia2.rl.agents package
      • Submodules
      • qualia2.rl.agents.ddqn module
      • qualia2.rl.agents.dqn module
      • qualia2.rl.agents.td3 module
      • Module contents
    • qualia2.rl.envs package
      • Submodules
      • qualia2.rl.envs.atari module
      • qualia2.rl.envs.box2d module
      • qualia2.rl.envs.classic_control module
      • qualia2.rl.envs.roboschool module
      • qualia2.rl.envs.toy_text module
      • Module contents
  • Submodules
  • qualia2.rl.memory module
  • qualia2.rl.rl_core module
  • qualia2.rl.rl_util module
  • Module contents
• qualia2.text package
  • Submodules
  • qualia2.text.cbow module
  • qualia2.text.text_util module
  • Module contents
• qualia2.vision package
  • Submodules
  • qualia2.vision.alexnet module
  • qualia2.vision.densenet module
  • qualia2.vision.imagenet_labels module
  • qualia2.vision.openpose module
  • qualia2.vision.resnet module
  • qualia2.vision.squeezenet module
  • qualia2.vision.transforms module
  • qualia2.vision.vgg module
  • Module contents

Submodules

qualia2.autograd module

class qualia2.autograd.Abs(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

calc_grad(dx)

calculates gradients for backpropagation

static forward(a)

calculates forward propagation

class qualia2.autograd.Add(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

Adds arrays elementwise.

calc_grad(dx)

calculates gradients for backpropagation

static forward(*args)

calculates forward propagation

class qualia2.autograd.Clamp(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

calc_grad(dx)

calculates gradients for backpropagation

static forward(x, low, high)

calculates forward propagation

class qualia2.autograd.Div(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

Elementwise true division

calc_grad(dx)

calculates gradients for backpropagation

static forward(a, b)

calculates forward propagation

class qualia2.autograd.Expand_dims(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

calc_grad(dx)

calculates gradients for backpropagation

static forward(a, axis)

calculates forward propagation

class qualia2.autograd.Function(output_shape, *args, **kwargs)

Bases: object

All function should inherit this class.

forward and calc_grad methods should be overwritten.

Attributes:

output_shape (tuple(int)): output shape of a function

var (tuple(Tensor)): Tensor(s) that was feeded

kwargs (dict): some useful data for backward calculation

backward(*args)

executes backpropagation

calc_grad(*args)

calculates gradients for backpropagation

static forward(*args, **kwargs)

calculates forward propagation

static handle_broadcast(arg, trg)

classmethod prepare(output_shape, *args, **kwargs)

class qualia2.autograd.Gather(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

Gathers values along an axis specified by dim.

calc_grad(dx)

calculates gradients for backpropagation

static forward(a, dim, idx)

calculates forward propagation

class qualia2.autograd.Matmul(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

calc_grad(dx)

calculates gradients for backpropagation

static forward(a, b)

calculates forward propagation

class qualia2.autograd.Mul(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

Multiplies two arrays elementwise.

calc_grad(dx)

calculates gradients for backpropagation

static forward(a, b)

calculates forward propagation

class qualia2.autograd.Neg(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

Takes numerical negative elementwise.

calc_grad(dx)

calculates gradients for backpropagation

static forward(a)

calculates forward propagation

class qualia2.autograd.Pow(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

Computes a ** b elementwise.

calc_grad(dx)

calculates gradients for backpropagation

static forward(a, b)

calculates forward propagation

class qualia2.autograd.Reshape(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

calc_grad(dx)

calculates gradients for backpropagation

static forward(a, shape)

calculates forward propagation

class qualia2.autograd.Slice(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

calc_grad(dx)

calculates gradients for backpropagation

static forward(a, slice)

calculates forward propagation

class qualia2.autograd.Squeeze(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

calc_grad(dx)

calculates gradients for backpropagation

static forward(a, axis=None)

calculates forward propagation

class qualia2.autograd.Sub(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

Subtracts arguments elementwise.

calc_grad(dx)

calculates gradients for backpropagation

static forward(*args)

calculates forward propagation

class qualia2.autograd.Tensor(data, requires_grad=True, dtype='float64')

Bases: object

Wrapper class to execute automatic differentiation

Args:

data (Union[ndarray,int,float]): tensor to compute the automatic differentiation

requires_grad (bool): Whether to store grads. If False is set, grad of the Tensor will be zeros. Default: True

dtype (str): data type of the tensor Default: ‘float64’

Attributes:

data (ndarray): Stores data of the Tensor

grad (ndarray): Stores gradients of the Tensor

creator (Function): Stores the creator of the Tensor, which will be called at the backpropagation.

child (list): temp list of child, which will be created in a forward pass

requires_grad (bool): Whether to store grads. If False is set, grad of the Tensor will be zeros.

shape (tuple): Stores the shape of Tensor’s data

ndim (int): Stores the number of Tensor’s data dimentions

Examples:

>>> # The following example will compute the dy/dx
>>> # Create Tensor objects 
>>> x = qualia2.array(5)
>>> # Write an equation 
>>> y = x**2 - 2*x + 1
>>> print(y)
>>> # Calclate gradiant 
>>> y.backward()
>>> # Print `dy/dx`
>>> print(x.grad)

property T

transpose of the Tensor

Returns:

(Tensor): transpose of the tensor

asnumpy()

aquire Tensor data as numpy ndarray

Returns:

(ndarray): numpy array

backward(*args)

calculates all the gradients in the graph

Args:

*args (ndarray): seed of the reverse accumulation AD; optional

clamp(low, high)

clamp the data

Args:

low (float): lower limit of the data.

high (float): upper limit of the data.

Returns:

(Tensor): clamped Tensor

copy(data)

detach()

returns a new Tensor, detached from the current graph.

Returns:

(Tensor): a new Tensor, detached from the current graph.

expand_dims(axis)

fill(val)

initialize the Tensor data with a constant value

Args:

val (float|int): a value to fill the Tensor data

gather(dim, idx)

gradasnumpy()

aquire Tensor grad as numpy ndarray

Returns:

(ndarray): numpy array

handle_const(obj)

handles the constant object such as int or float

Args:

obj (Union[Tensor,int,float]): constant

Returns:

(Tensor)

normal(mean=0, std=1)

initialize the Tensor data with normal distribution

Args:

mean (float): mean of the normal distribution.

std (float): std of the normal distribution.

ones()

initialize the Tensor data with ones

register_hook(hook)

reshape(*args)

set_creator(obj)

sets the creator of the Tensor

Args:

obj (Function): the function that created the Tensor

squeeze(axis=None)

transpose(*args)

uniform(low=0, high=1)

initialize the Tensor data with uniform distribution

Args:

low (float): lower limit of the uniform distribution.

high (float): upper limit of the uniform distribution.

unsqueeze(axis)

zeros()

initialize the Tensor data with zeros

class qualia2.autograd.Transpose(output_shape, *args, **kwargs)

Bases: qualia2.autograd.Function

calc_grad(dx)

calculates gradients for backpropagation

static forward(a, axes)

calculates forward propagation

qualia2.core module

qualia2.core.to_cpu(obj)

qualia2.core.to_gpu(obj)

qualia2.util module

class qualia2.util.Trainer(batch, path=None)

Bases: object

Trainer base class

after_episode(epoch, model, loss)

after_train()

before_episode(dataloader, model)

before_train(dataloader, model, filename)

plot(filename=None)

set_data_transformer(fn)

set_label_transformer(fn)

test(model, dataloader, batch, filename=None)

train(model, dataloader, optim, criterion, epochs=200, filename=None)

trainer helps the training process of supervised learning Args:

model (Module): model to train dataloader (DataLoader): dataloader to use optim (Optimizer): optimizer to use criterion (Function): loss function to use epochs (int): number of epochs filename (string): specify the filename as well as the loading path without the file extension. (ex) path/to/filename

train_routine(model, dataloader, optim, criterion, epochs=200)

qualia2.util.check_function(fn, *args, x=None, domain=(-1000.0, 1000.0), **kwargs)

qualia2.util.download_progress(count, block_size, total_size)

qualia2.util.numerical_grad(fn, tensor, *args, **kwargs)

qualia2.util.progressbar(progress, process, text_before='', text_after='')

Module contents

qualia2.arange(*args, dtype='float64')

qualia2.array(obj, dtype='float64')

qualia2.copy(tensor)

qualia2.empty(shape, dtype='float64')

qualia2.empty_like(obj, dtype='float64')

qualia2.linspace(start, stop, num=50, endpoint=True, retstep=False, dtype='float64')

qualia2.load(filename)

Loads parameters saved in qla format.

Args:

filename (str): specify the filename as well as the path to the file without the file extension. (ex) path/to/filename

qualia2.normal(mean=0, std=1, shape=None, dtype='float64')

qualia2.normal_like(obj, mean=0, std=1, dtype='float64')

qualia2.numel(obj)

qualia2.ones(shape, dtype='int64')

qualia2.ones_like(obj, dtype='int64')

qualia2.rand(*args, dtype='float64')

qualia2.rand_like(obj, dtype='float64')

qualia2.randint(low, high=None, shape=None, dtype='float64')

qualia2.randint_like(obj, low, high=None, dtype='float64')

qualia2.randn(*args, dtype='float64')

qualia2.randn_like(obj, dtype='float64')

qualia2.save(state_dict, filename, protocol=-1)

Saves parameters to qla format.

Args:

state_dict (dict): state dict to save filename (str): specify the filename as well as the saving path without the file extension. (ex) path/to/filename protocol (int): pickle protocol

qualia2.seed(seed=None)

qualia2.uniform(low=0.0, high=1.0, shape=None, dtype='float64')

qualia2.uniform_like(obj, low=0.0, high=1.0, dtype='float64')

qualia2.zeros(shape, dtype='int64')

qualia2.zeros_like(obj, dtype='int64')