thinc API reference

101 public APIs from thinc (explosion/thinc) — 22 classes, 41 functions, 38 methods. Signatures extracted by static analysis of the actual source.

Repository: explosion/thinc

Kind	Count
Classes	22
Functions	41
Methods	38

API list

classthinc.backends._param_server.ParamServer

Serve parameters for a single process.

functhinc.backends.get_current_ops() -> Ops

Get the current backend object.

functhinc.backends.get_ops(name:str, **kwargs) -> Ops

Get a backend object.

classthinc.backends.mps_ops.MPSOps

Ops class for Metal Performance shaders.

functhinc.backends.ops.gaussian_cdf(ops:Ops, X:FloatsXdT) -> FloatsXdT

Gaussian CDF for distribution with mean 0 and stdev 1.

functhinc.backends.ops.gaussian_pdf(ops:Ops, X:FloatsXdT) -> FloatsXdT

Gaussian PDF for distribution with mean 0 and stdev 1.

functhinc.backends.set_current_ops(ops:Ops) -> None

Change the current backend object.

functhinc.backends.set_gpu_allocator(allocator:str) -> None

Route GPU memory allocation via PyTorch or tensorflow.

functhinc.backends.use_pytorch_for_gpu_memory() -> None

Route GPU memory allocation via PyTorch.

functhinc.backends.use_tensorflow_for_gpu_memory() -> None

Route GPU memory allocation via TensorFlow.

methodthinc.config.registry.create(registry_name:str, entry_points:bool=False) -> None

Create a new custom registry.

functhinc.layers.array_getitem.array_getitem(index:Index) -> Model[ArrayTXd, ArrayTXd]

Index into input arrays, and return the subarrays.

functhinc.layers.array_getitem.floats_getitem(index:Index) -> Model[FloatsXd, FloatsXd]

Index into input arrays, and return the subarrays.

functhinc.layers.array_getitem.ints_getitem(index:Index) -> Model[IntsXd, IntsXd]

Index into input arrays, and return the subarrays.

functhinc.layers.bidirectional.bidirectional(l2r:Model[InT, OutT], r2l:Optional[Model[InT, OutT]]=None) -> Model[InT, OutT]

Stitch two RNN models into a bidirectional layer.

functhinc.layers.clone.clone(orig:Model[InT, OutT], n:int) -> Model[InT, OutT]

Construct `n` copies of a layer, with distinct weights.

functhinc.layers.map_list.map_list(layer:Model[InT, OutT]) -> Model[List[InT], List[OutT]]

Create a model that maps a child layer across list inputs.

func

thinc.layers.mish.Mish(nO:Optional[int]=None, nI:Optional[int]=None, *init_W:Optional[Callable]=None, *init_b:Optional[Callable]=None, *dropout:Optional[float]=None, *normalize:bool=False) -> Model[InT, OutT]

Dense layer with mish activation.

functhinc.layers.noop.noop(*layers:Model) -> Model[InOutT, InOutT]

Transform a sequences of layers into a null operation.

functhinc.layers.parametricattention.ParametricAttention(nO:Optional[int]=None) -> Model[InT, OutT]

Weight inputs by similarity to a learned vector

functhinc.layers.ragged2list.ragged2list() -> Model[InT, OutT]

Transform sequences from a ragged format into lists.

functhinc.layers.resizable.resizable(layer, resize_layer:Callable) -> Model[InT, OutT]

Container that holds one layer that can change dimensions.

functhinc.layers.with_reshape.with_reshape(layer:Model[OutT, OutT]) -> Model[InT, InT]

Reshape data on the way into and out from a layer.

classthinc.loss.Loss

Base class for classes computing the loss / gradient.

methodthinc.loss.Loss.get_grad(guesses:GuessT, truths:TruthT) -> GradT

Get the gradient of the loss.

classthinc.model.Model

Class for implementing Thinc models and layers.

methodthinc.model.Model.attrs() -> Dict[str, Any]

A dict of the model's attrs.

methodthinc.model.Model.can_from_bytes(bytes_data:bytes, *strict:bool=True) -> bool

Check whether the bytes data is compatible with the model.

methodthinc.model.Model.can_from_dict(msg:Dict, *strict:bool=True) -> bool

Check whether a dictionary is compatible with the model.

methodthinc.model.Model.copy() -> SelfT

Create a copy of the model, its attributes, and its parameters.

methodthinc.model.Model.dim_names() -> Tuple[str, ...]

Get the names of registered dimensions (including unset).

methodthinc.model.Model.finish_update(optimizer:Optimizer) -> None

Update parameters with current gradients.

methodthinc.model.Model.from_bytes(bytes_data:bytes) -> 'Model'

Deserialize the model from a bytes representation.

methodthinc.model.Model.from_disk(path:Union[Path, str]) -> 'Model'

Deserialize the model from disk.

methodthinc.model.Model.get_dim(name:str) -> int

Retrieve the value of a dimension of the given name.

methodthinc.model.Model.get_grad(name:str) -> FloatsXd

Get a gradient from the model.

methodthinc.model.Model.get_param(name:str) -> FloatsXd

Retrieve a weights parameter by name.

methodthinc.model.Model.get_ref(name:str) -> 'Model'

Retrieve the value of a reference of the given name.

methodthinc.model.Model.has_dim(name:str) -> Optional[bool]

Check whether the model has a dimension of a given name.

methodthinc.model.Model.has_grad(name:str) -> bool

Check whether the model has a non-zero gradient for a parameter.

methodthinc.model.Model.has_param(name:str) -> Optional[bool]

Check whether the model has a weights parameter of the given name.

methodthinc.model.Model.has_ref(name:str) -> Optional[bool]

Check whether the model has a reference of a given name.

methodthinc.model.Model.inc_grad(name:str, value:FloatsXd) -> None

Increment the gradient of a parameter by a value.

methodthinc.model.Model.layers() -> List['Model']

A list of child layers of the model.

methodthinc.model.Model.maybe_get_dim(name:str) -> Optional[int]

Retrieve the value of a dimension of the given name, or None.

methodthinc.model.Model.maybe_get_grad(name:str) -> Optional[FloatsXd]

Retrieve a gradient by name, or None.

methodthinc.model.Model.maybe_get_param(name:str) -> Optional[FloatsXd]

Retrieve a weights parameter by name, or None.

methodthinc.model.Model.maybe_get_ref(name:str) -> Optional['Model']

Retrieve the value of a reference if it exists, or None.

methodthinc.model.Model.param_names() -> Tuple[str, ...]

Get the names of registered parameter (including unset).

methodthinc.model.Model.remove_node(node:'Model') -> None

Remove a node from all layers lists, and then update references.

methodthinc.model.Model.replace_node(old:'Model', new:'Model') -> bool

Replace a node anywhere it occurs within the model.

methodthinc.model.Model.set_dim(name:str, value:int, *force:bool=False) -> None

Set a value for a dimension.

methodthinc.model.Model.set_grad(name:str, value:FloatsXd) -> None

Set a gradient value for the model.

methodthinc.model.Model.set_param(name:str, value:Optional[FloatsXd]) -> None

Set a weights parameter's value.

methodthinc.model.Model.set_ref(name:str, value:Optional['Model']) -> None

Set a value for a reference.

methodthinc.model.Model.to_bytes() -> bytes

Serialize the model to a bytes representation.

methodthinc.model.Model.to_cpu() -> None

Transfer the model to CPU.

methodthinc.model.Model.to_dict() -> Dict

Serialize the model to a dict representation.

methodthinc.model.Model.to_disk(path:Union[Path, str]) -> None

Serialize the model to disk.

methodthinc.model.Model.to_gpu(gpu_id:int) -> None

Transfer the model to a given GPU device.

methodthinc.model.Model.walk(*order:str='bfs') -> Iterable['Model']

Iterate out layers of the model.

functhinc.model.deserialize_attr(_:Any, value:Any, name:str, model:Model) -> Any

Deserialize an attribute value (defaults to msgpack).

functhinc.model.serialize_attr(_:Any, value:Any, name:str, model:Model) -> bytes

Serialize an attribute value (defaults to msgpack).

functhinc.model.set_dropout_rate(model:_ModelT, drop:float, attrs=['dropout_rate']) -> _ModelT

Walk over the model's nodes, setting the dropout rate.

functhinc.model.wrap_model_recursive(model:Model, wrapper:Callable[[Model], _ModelT]) -> _ModelT

Recursively wrap a model and its submodules.

functhinc.mypy.get_reducers_type(ctx:FunctionContext) -> Type

Determine a more specific model type for functions that combine models.

functhinc.schedules.compounding(start:float, stop:float, compound:float, *t:float=0.0) -> Iterable[float]

Yield an infinite series of compounding values.

functhinc.schedules.constant(rate:float) -> Iterable[float]

Yield a constant rate.

classthinc.shims.mxnet.MXNetShim

Interface between a MXNet model and a Thinc Model.

classthinc.shims.pytorch.PyTorchShim

Interface between a PyTorch model and a Thinc Model.

functhinc.shims.pytorch.default_serialize_torch_model(model:Any) -> bytes

Serializes the parameters of the wrapped PyTorch model to bytes.

classthinc.shims.pytorch_grad_scaler.PyTorchGradScaler

Gradient scaler for the PyTorch shim.

methodthinc.shims.pytorch_grad_scaler.PyTorchGradScaler.unscale(tensors)

Unscale the given tensors.

classthinc.shims.shim.Shim

Define a basic interface for external models.

classthinc.shims.tensorflow.TensorFlowShim

Interface between a TensorFlow model and a Thinc Model.

classthinc.shims.torchscript.TorchScriptShim

A Thinc shim that wraps a TorchScript module.

classthinc.types.Floats1d

1-dimensional array of floats.

classthinc.types.Floats2d

2-dimensional array of floats

classthinc.types.Floats3d

3-dimensional array of floats

classthinc.types.Floats4d

4-dimensional array of floats.

classthinc.types.Generator

Custom generator type.

classthinc.types.Ints1d

1-dimensional array of ints.

classthinc.types.Ints2d

2-dimensional array of ints.

classthinc.types.Ints3d

3-dimensional array of ints.

classthinc.types.Ints4d

4-dimensional array of ints.

classthinc.types.Padded

A batch of padded sequences, sorted by decreasing length.

classthinc.types.Unserializable

Wrap a value to prevent it from being serialized by msgpack.

classthinc.util.ArrayInfo

Container for info for checking array compatibility.

functhinc.util.assert_mxnet_installed() -> None

Raise an ImportError if MXNet is not installed.

functhinc.util.assert_pytorch_installed() -> None

Raise an ImportError if PyTorch is not installed.

functhinc.util.assert_tensorflow_installed() -> None

Raise an ImportError if TensorFlow is not installed.

functhinc.util.fix_random_seed(seed:int=0) -> None

Set the random seed across random, numpy.random and cupy.random.

functhinc.util.is_cupy_array(obj:Any) -> bool

Check whether an object is a cupy array.

functhinc.util.is_numpy_array(obj:Any) -> bool

Check whether an object is a numpy array.

functhinc.util.is_xp_array(obj:Any) -> bool

Check whether an object is a numpy or cupy array.

functhinc.util.mxnet2xp(mx_tensor:'mx.nd.NDArray', *ops:Optional['Ops']=None) -> ArrayXd

Convert a MXNet tensor to a numpy or cupy tensor.

functhinc.util.prefer_gpu(gpu_id:int=0) -> bool

Use GPU if it's available.

functhinc.util.require_cpu() -> bool

Use CPU through best available backend.

functhinc.util.set_active_gpu(gpu_id:int) -> 'cupy.cuda.Device'

Set the current GPU device for cupy and torch (if available).

functhinc.util.xp2mxnet(xp_tensor:ArrayXd, requires_grad:bool=False) -> 'mx.nd.NDArray'

Convert a numpy or cupy tensor to a MXNet tensor.

functhinc.util.xp2torch(xp_tensor:ArrayXd, requires_grad:bool=False, device:Optional['torch.device']=None) -> 'torch.Tensor'

Convert a numpy or cupy tensor to a PyTorch tensor.

About this data

These signatures were extracted from the public source of explosion/thinc using Python's ast module. Argument names, default values, type annotations and return types are taken verbatim from the code. Implementation bodies are never stored. See how it works for details.

Back to all 805 libraries