Skip to main content

Open API to/fro routes, models, and tests. Convert between docstrings, classes, methods, argparse, and SQLalchemy.

Project description

cdd-python

Python version range Python implementation License Linting, testing, coverage, and release Tested OSs, others may work Documentation coverage codecov black Imports: isort PyPi: release

Open API to/fro routes, models, and tests. Convert between docstrings, classes, methods, argparse, and SQLalchemy.

Public SDK works with filenames, source code, and even in memory constructs (e.g., as imported into your REPL).

Install package

PyPi

pip install python-cdd

Developer

pip install -r https://raw.githubusercontent.com/offscale/cdd-python/master/requirements.txt
pip install https://api.github.com/repos/offscale/cdd-python/zipball#egg=cdd

Purpose

This was created to aid in the ml_params project. It exposes an @abstractclass which is implemented [officially] by more than 8 projects.

Its def train(self, <these>) has a potentially large number of arguments. Additionally, there is a def train_c(self, config), which accepts an instance of a Config class, or a dictionary. Finally: ml_params defines a CLI interface.

With current tooling there is no way to know:

  • What arguments can be provided to train
  • What CLI arguments are available
  • What 'shape' the Config takes

Some of these problems can be solved dynamically, however in doing so one loses developer-tool insights. There is no code-completion, and likely the CLI parser won't provide you with the enumeration of possibilities.

Example (REPL)

To create a class from tf.keras.optimizers.Adam:

>>> from cdd.source_transformer import to_code

>>> from cdd import emit, parse

>>> import tensorflow as tf

>>> from typing import Optional

>>> print(to_code(emit.class_(parse.class_(tf.keras.optimizers.Adam,
                                           merge_inner_function="__init__"),
                              class_name="AdamConfig")))


class AdamConfig(object):
    """
    Optimizer that implements the Adam algorithm.

    Adam optimization is a stochastic gradient descent method that is based on
    adaptive estimation of first-order and second-order moments.

    According to
    [Kingma et al., 2014](http://arxiv.org/abs/1412.6980),
    the method is "*computationally
    efficient, has little memory requirement, invariant to diagonal rescaling of
    gradients, and is well suited for problems that are large in terms of
    data/parameters*".


    Usage:

    >>> opt = tf.keras.optimizers.Adam(learning_rate=0.1)
    >>> var1 = tf.Variable(10.0)
    >>> loss = lambda: (var1 ** 2)/2.0       # d(loss)/d(var1) == var1
    >>> step_count = opt.minimize(loss, [var1]).numpy()
    >>> # The first step is `-learning_rate*sign(grad)`
    >>> var1.numpy()
    9.9

    Reference:
      - [Kingma et al., 2014](http://arxiv.org/abs/1412.6980)
      - [Reddi et al., 2018](
          https://openreview.net/pdf?id=ryQu7f-RZ) for `amsgrad`.

    Notes:

    The default value of 1e-7 for epsilon might not be a good default in
    general. For example, when training an Inception network on ImageNet a
    current good choice is 1.0 or 0.1. Note that since Adam uses the
    formulation just before Section 2.1 of the Kingma and Ba paper rather than
    the formulation in Algorithm 1, the "epsilon" referred to here is "epsilon
    hat" in the paper.

    The sparse implementation of this algorithm (used when the gradient is an
    IndexedSlices object, typically because of `tf.gather` or an embedding
    lookup in the forward pass) does apply momentum to variable slices even if
    they were not used in the forward pass (meaning they have a gradient equal
    to zero). Momentum decay (beta1) is also applied to the entire momentum
    accumulator. This means that the sparse behavior is equivalent to the dense
    behavior (in contrast to some momentum implementations which ignore momentum
    unless a variable slice was actually used).

    :cvar learning_rate: A `Tensor`, floating point value, or a schedule that is a
        `tf.keras.optimizers.schedules.LearningRateSchedule`, or a callable that takes no arguments and
        returns the actual value to use, The learning rate.
    :cvar beta_1: A float value or a constant float tensor, or a callable that takes no arguments and
        returns the actual value to use. The exponential decay rate for the 1st moment estimates.
    :cvar beta_2: A float value or a constant float tensor, or a callable that takes no arguments and
        returns the actual value to use, The exponential decay rate for the 2nd moment estimates.
    :cvar epsilon: A small constant for numerical stability. This epsilon is "epsilon hat" in the
        Kingma and Ba paper (in the formula just before Section 2.1), not the epsilon in Algorithm 1 of the
        paper.
    :cvar amsgrad: Boolean. Whether to apply AMSGrad variant of this algorithm from the paper "On the
        Convergence of Adam and beyond".
    :cvar name: Optional name for the operations created when applying gradients.
    :cvar kwargs: Keyword arguments. Allowed to be one of `"clipnorm"` or `"clipvalue"`. `"clipnorm"`
        (float) clips gradients by norm; `"clipvalue"` (float) clips gradients by value."""
    learning_rate: float = 0.001
    beta_1: float = 0.9
    beta_2: float = 0.999
    epsilon: float = 1e-07
    amsgrad: bool = False
    name: Optional[str] = 'Adam'
    kwargs: Optional[dict] = None
    _HAS_AGGREGATE_GRAD: bool = True

Approach

Traverse the AST, and emit the modifications, such that each of these 4 can convert to each other.

Docstring

Acquire from the official tensorflow_datasets model zoo, or the ophthalmology focussed ml-prepare library

:param dataset_name: name of dataset. Defaults to mnist
:type dataset_name: ```str```

:param tfds_dir: directory to look for models in. Defaults to ~/tensorflow_datasets
:type tfds_dir: ```Optional[str]```

:param K: backend engine, e.g., `np` or `tf`. Defaults to np
:type K: ```Union[np, tf]```

:param as_numpy: Convert to numpy ndarrays
:type as_numpy: ```Optional[bool]```

:param data_loader_kwargs: pass this as arguments to data_loader function
:type data_loader_kwargs: ```**data_loader_kwargs```

:return: Train and tests dataset splits. Defaults to (np.empty(0), np.empty(0))
:rtype: ```Union[Tuple[tf.data.Dataset, tf.data.Dataset], Tuple[np.ndarray, np.ndarray]]```
class
from typing import Optional, Union, Tuple, Literal

import numpy as np
import tensorflow as tf


class TargetClass(object):
    """
    Acquire from the official tensorflow_datasets model zoo, or the ophthalmology focussed ml-prepare library

    :cvar dataset_name: name of dataset. Defaults to mnist
    :cvar tfds_dir: directory to look for models in. Defaults to ~/tensorflow_datasets
    :cvar K: backend engine, e.g., `np` or `tf`. Defaults to np
    :cvar as_numpy: Convert to numpy ndarrays
    :cvar data_loader_kwargs: pass this as arguments to data_loader function
    :cvar return_type: Train and tests dataset splits. Defaults to (np.empty(0), np.empty(0))"""

    dataset_name: str = 'mnist'
    tfds_dir: Optional[str] = '~/tensorflow_datasets'
    K: Literal['np', 'tf'] = 'np'
    as_numpy: Optional[bool] = None
    data_loader_kwargs: dict = {}
    return_type: Union[Tuple[tf.data.Dataset, tf.data.Dataset], Tuple[np.ndarray, np.ndarray]] = (
        np.empty(0),
        np.empty(0),
    )
class method
from typing import Optional, Union, Tuple, Literal

import numpy as np
import tensorflow as tf

class C(object):
    """ C class (mocked!) """

    def method_name(
        self,
        dataset_name: str = 'mnist',
        tfds_dir: Optional[str] = '~/tensorflow_datasets',
        K: Literal['np', 'tf'] = 'np',
        as_numpy: Optional[bool] = None,
        **data_loader_kwargs
    ) -> Union[Tuple[tf.data.Dataset, tf.data.Dataset], Tuple[np.ndarray, np.ndarray]]:
        """
        Acquire from the official tensorflow_datasets model zoo, or the ophthalmology focussed ml-prepare library

        :param dataset_name: name of dataset.

        :param tfds_dir: directory to look for models in.

        :param K: backend engine, e.g., `np` or `tf`.

        :param as_numpy: Convert to numpy ndarrays

        :param data_loader_kwargs: pass this as arguments to data_loader function

        :return: Train and tests dataset splits.
        """
        return np.empty(0), np.empty(0)
Argparse augmenting function
from typing import Union, Tuple
from json import loads

import numpy as np
import tensorflow as tf


def set_cli_args(argument_parser):
    """
    Set CLI arguments

    :param argument_parser: argument parser
    :type argument_parser: ```ArgumentParser```

    :return: argument_parser, Train and tests dataset splits.
    :rtype: ```Tuple[ArgumentParser, Union[Tuple[tf.data.Dataset, tf.data.Dataset], Tuple[np.ndarray, np.ndarray]]]```
    """
    argument_parser.description = (
        'Acquire from the official tensorflow_datasets model zoo, or the ophthalmology focussed ml-prepare library'
    )
    argument_parser.add_argument(
        '--dataset_name', type=str, help='name of dataset.', required=True, default='mnist'
    )
    argument_parser.add_argument(
        '--tfds_dir',
        type=str,
        help='directory to look for models in.',
        default='~/tensorflow_datasets',
    )
    argument_parser.add_argument(
        '--K',
        type=globals().__getitem__,
        choices=('np', 'tf'),
        help='backend engine, expr.g., `np` or `tf`.',
        required=True,
        default='np',
    )
    argument_parser.add_argument('--as_numpy', type=bool, help='Convert to numpy ndarrays')
    argument_parser.add_argument(
        '--data_loader_kwargs', type=loads, help='pass this as arguments to data_loader function'
    )
    return argument_parser, (np.empty(0), np.empty(0))
SQLalchemy

There are two variants in the latest SQLalchemy, both are supported:

from sqlalchemy import JSON, Boolean, Column, Enum, MetaData, String, Table, create_engine

engine = create_engine("sqlite://", echo=True, future=True)
metadata = MetaData()

config_tbl = Table(
    "config_tbl",
    metadata,
    Column(
        "dataset_name",
        String,
        doc="name of dataset",
        default="mnist",
        primary_key=True,
    ),
    Column(
        "tfds_dir",
        String,
        doc="directory to look for models in",
        default="~/tensorflow_datasets",
        nullable=False,
    ),
    Column(
        "K",
        Enum("np", "tf", name="K"),
        doc="backend engine, e.g., `np` or `tf`",
        default="np",
        nullable=False,
    ),
    Column(
        "as_numpy",
        Boolean,
        doc="Convert to numpy ndarrays",
        default=None,
        nullable=True,
    ),
    Column(
        "data_loader_kwargs",
        JSON,
        doc="pass this as arguments to data_loader function",
        default=None,
        nullable=True,
    ),
    comment='Acquire from the official tensorflow_datasets model zoo, or the ophthalmology focussed ml-prepare\n'
            '\n'
            ':returns: Train and tests dataset splits. Defaults to (np.empty(0), np.empty(0))\n'
            ':rtype: ```Union[Tuple[tf.data.Dataset, tf.data.Dataset], Tuple[np.ndarray, np.ndarray]]```',
)

metadata.create_all(engine)
from sqlalchemy.orm import declarative_base
from sqlalchemy import JSON, Boolean, Column, Enum, String

Base = declarative_base()

class Config(Base):
    """
    Acquire from the official tensorflow_datasets model zoo, or the ophthalmology focussed ml-prepare

    :returns: Train and tests dataset splits. Defaults to (np.empty(0), np.empty(0))
    :rtype: ```Union[Tuple[tf.data.Dataset, tf.data.Dataset], Tuple[np.ndarray, np.ndarray]]```
    """
    __tablename__ = "config_tbl"

    dataset_name = Column(
        String,
        doc="name of dataset",
        default="mnist",
        primary_key=True,
    )

    tfds_dir = Column(
        String,
        doc="directory to look for models in",
        default="~/tensorflow_datasets",
        nullable=False,
    )

    K = Column(
        Enum("np", "tf", name="K"),
        doc="backend engine, e.g., `np` or `tf`",
        default="np",
        nullable=False,
    )

    as_numpy = Column(
        Boolean,
        doc="Convert to numpy ndarrays",
        default=None,
        nullable=True,
    )

    data_loader_kwargs = Column(
        JSON,
        doc="pass this as arguments to data_loader function",
        default=None,
        nullable=True,
    )

    def __repr__(self):
        """
        Emit a string representation of the current instance

        :returns: String representation of instance
        :rtype: ```str```
        """

        return ("Config(dataset_name={dataset_name!r}, tfds_dir={tfds_dir!r}, "
                "K={K!r}, as_numpy={as_numpy!r}, data_loader_kwargs={data_loader_kwargs!r})").format(
            dataset_name=self.dataset_name, tfds_dir=self.tfds_dir, K=self.K,
            as_numpy=self.as_numpy, data_loader_kwargs=self.data_loader_kwargs
        )

Advantages

  • CLI gives proper --help messages
  • IDE and console gives proper insights to function, and arguments, including on type
  • class–based interface opens this up to clean object passing
  • @abstractmethod can add—remove, and change—as many arguments as it wants; including required arguments; without worry
  • Verbosity of output removes the magic. It's always clear what's going on.
  • Outputting regular code means things can be composed and extended as normally.

Disadvantages

  • You have to run a tool to synchronise your: docstring(s), config class(es), and argparse augmenting function.
  • Duplication (but the tool handles this)

Alternatives

  • Dynamic code generation, e.g., with a singular interface for everything; so everything is in one place without duplication

Minor other use-cases this facilitates

  • Switch between having types in the docstring and having the types inline (PEP484–style))
  • Switch between docstring formats (to/from {numpy, ReST, google})
  • Desktop GUI with wxWidgets, from the argparse layer through Gooey [one liner]

CLI for this project

$ python -m cdd --help

usage: python -m cdd [-h] [--version] {sync_properties,sync,gen} ...

Translate between docstrings, classes, methods, and argparse.

positional arguments:
  {sync_properties,sync,gen}
    sync_properties     Synchronise one or more properties between input and
                        input_str Python files
    sync                Force argparse, classes, and/or methods to be
                        equivalent
    gen                 Generate classes, functions, and/or argparse functions
                        from the input mapping

optional arguments:
  -h, --help            show this help message and exit
  --version             show program's version number and exit

sync

$ python -m cdd sync --help

usage: python -m cdd sync [-h] [--argparse-function ARGPARSE_FUNCTIONS]
                               [--argparse-function-name ARGPARSE_FUNCTION_NAMES]
                               [--class CLASSES] [--class-name CLASS_NAMES]
                               [--function FUNCTIONS]
                               [--function-name FUNCTION_NAMES] --truth
                               {argparse_function,class,function}

optional arguments:
  -h, --help            show this help message and exit
  --argparse-function ARGPARSE_FUNCTIONS
                        File where argparse function is `def`ined.
  --argparse-function-name ARGPARSE_FUNCTION_NAMES
                        Name of argparse function.
  --class CLASSES       File where class `class` is declared.
  --class-name CLASS_NAMES
                        Name of `class`
  --function FUNCTIONS  File where function is `def`ined.
  --function-name FUNCTION_NAMES
                        Name of Function. If method, use Python resolution
                        syntax, i.e., ClassName.function_name
  --truth {argparse_function,class,function}
                        Single source of truth. Others will be generated from
                        this. Will run with first found choice.

sync_properties

$ python -m cdd sync_properties --help

usage: python -m cdd sync_properties [-h] --input-filename INPUT_FILENAME
                                          --input-param INPUT_PARAMS
                                          [--input-eval] --output-filename
                                          OUTPUT_FILENAME --output-param
                                          OUTPUT_PARAMS
                                          [--output-param-wrap OUTPUT_PARAM_WRAP]

optional arguments:
  -h, --help            show this help message and exit
  --input-filename INPUT_FILENAME
                        File to find `--input-param` from
  --input-param INPUT_PARAMS
                        Location within file of property. Can be top level
                        like `a` for `a=5` or with the `.` syntax as in
                        `--output-param`.
  --input-eval          Whether to evaluate the input-param, or just leave it
  --output-filename OUTPUT_FILENAME
                        Edited in place, the property within this file (to
                        update) is selected by --output-param
  --output-param OUTPUT_PARAMS
                        Parameter to update. E.g., `A.F` for `class A: F`,
                        `f.g` for `def f(g): pass`
  --output-param-wrap OUTPUT_PARAM_WRAP
                        Wrap all input_str params with this. E.g.,
                        `Optional[Union[{output_param}, str]]`

gen

$ python -m cdd gen --help

usage: python -m cdd gen [-h] --name-tpl NAME_TPL --input-mapping
                              INPUT_MAPPING [--prepend PREPEND]
                              [--imports-from-file IMPORTS_FROM_FILE] --type
                              {argparse,class,function} --output-filename
                              OUTPUT_FILENAME

optional arguments:
  -h, --help            show this help message and exit
  --name-tpl NAME_TPL   Template for the name, e.g., `{name}Config`.
  --input-mapping INPUT_MAPPING
                        Import location of dictionary/mapping/2-tuple
                        collection.
  --prepend PREPEND     Prepend file with this. Use '\n' for newlines.
  --imports-from-file IMPORTS_FROM_FILE
                        Extract imports from file and append to `output_file`.
                        If module or other symbol path given, resolve file
                        then use it.
  --type {argparse,class,function}
                        What type to generate.
  --output-filename OUTPUT_FILENAME, -o OUTPUT_FILENAME
                        Output file to write to.

License

Licensed under either of

at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.

Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

python-cdd-0.0.66a2.tar.gz (139.0 kB view hashes)

Uploaded Source

Built Distribution

python_cdd-0.0.66a2-py3-none-any.whl (161.4 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page