Test your AI Agents with the all-new Agent to Agent Testing Platform.Learn More

How to use inner method in autotest

Best Python code snippet using autotest_python

meta.py

Source:meta.py

1# global2import ivy3from ivy.functional.ivy.core.gradients import gradient_descent_update4# Private #5# --------#6def _compute_cost_and_update_grads(cost_fn, order, batch, variables, outer_v, keep_outer_v,7                                   average_across_steps_or_final, all_grads, unique_outer, batched, num_tasks):8    if order == 1:9        cost, inner_grads = ivy.execute_with_gradients(10            lambda v: cost_fn(batch, v=variables.set_at_key_chains(v) if unique_outer else v),11            variables.at_key_chains(outer_v, ignore_none=True) if keep_outer_v else12            variables.prune_key_chains(outer_v, ignore_none=True), retain_grads=False)13        if batched:14            inner_grads = inner_grads * num_tasks15        if average_across_steps_or_final:16            all_grads.append(inner_grads)17    else:18        cost = cost_fn(batch, v=variables)19    return cost20def _train_task(inner_batch, outer_batch, inner_cost_fn, outer_cost_fn, variables, inner_grad_steps,21                inner_learning_rate, inner_optimization_step, order, average_across_steps, inner_v, keep_innver_v,22                outer_v, keep_outer_v, batched, num_tasks, stop_gradients):23    # init24    total_cost = 025    all_grads = list()26    # inner and outer27    unique_inner = inner_v is not None28    unique_outer = outer_v is not None29    # iterate through inner loop training steps30    for i in range(inner_grad_steps):31        # compute inner gradient for update the inner variables32        cost, inner_update_grads = ivy.execute_with_gradients(33            lambda v: inner_cost_fn(inner_batch, v=variables.set_at_key_chains(v) if unique_inner else v),34            variables.at_key_chains(inner_v, ignore_none=True) if keep_innver_v else35            variables.prune_key_chains(inner_v, ignore_none=True), retain_grads=order > 1)36        if batched:37            inner_update_grads = inner_update_grads * num_tasks38        # compute the cost to be optimized, and update all_grads if fist order method39        if outer_cost_fn is None and not unique_inner and not unique_outer:40            all_grads.append(inner_update_grads)41        else:42            cost = _compute_cost_and_update_grads(43                inner_cost_fn if outer_cost_fn is None else outer_cost_fn, order, outer_batch, variables, outer_v,44                keep_outer_v, average_across_steps, all_grads, unique_outer, batched, num_tasks)45        # update cost and update parameters46        total_cost = total_cost + cost47        if unique_inner:48            variables = variables.set_at_key_chains(49                inner_optimization_step(variables.at_key_chains(inner_v) if keep_innver_v else50                                        variables.prune_key_chains(inner_v), inner_update_grads,51                                        inner_learning_rate, inplace=False, stop_gradients=stop_gradients))52        else:53            variables = inner_optimization_step(variables, inner_update_grads, inner_learning_rate, inplace=False,54                                                stop_gradients=stop_gradients)55    # once training is finished, compute the final cost, and update all_grads if fist order method56    final_cost = _compute_cost_and_update_grads(57        inner_cost_fn if outer_cost_fn is None else outer_cost_fn, order, outer_batch, variables, outer_v,58        keep_outer_v, True, all_grads, unique_outer, batched, num_tasks)59    # update variables60    if stop_gradients:61        variables = variables.stop_gradients()62    if not batched:63        variables = variables.expand_dims(0)64    # average the cost or gradients across all timesteps if this option is chosen65    if average_across_steps:66        total_cost = total_cost + final_cost67        if order == 1:68            all_grads = sum(all_grads) / max(len(all_grads), 1)69        return total_cost / (inner_grad_steps + 1), variables, all_grads70    # else return only the final values71    if order == 1:72        all_grads = all_grads[-1]73    return final_cost, variables, all_grads74def _train_tasks_batched(batch, inner_batch_fn, outer_batch_fn, inner_cost_fn, outer_cost_fn, variables,75                         inner_grad_steps, inner_learning_rate, inner_optimization_step, order, average_across_steps,76                         inner_v, keep_innver_v, outer_v, keep_outer_v, return_inner_v, num_tasks, stop_gradients):77    inner_batch = batch78    outer_batch = batch79    if inner_batch_fn is not None:80        inner_batch = inner_batch_fn(inner_batch)81    if outer_batch_fn is not None:82        outer_batch = outer_batch_fn(outer_batch)83    cost, updated_ivs, grads = _train_task(inner_batch, outer_batch, inner_cost_fn, outer_cost_fn, variables,84                                           inner_grad_steps, inner_learning_rate, inner_optimization_step, order,85                                           average_across_steps, inner_v, keep_innver_v, outer_v, keep_outer_v, True,86                                           num_tasks, stop_gradients)87    grads = grads.reduce_mean(0) if isinstance(grads, ivy.Container) else grads88    if order == 1:89        if return_inner_v in ['all', True]:90            return cost, grads, updated_ivs91        elif return_inner_v == 'first':92            return cost, grads, updated_ivs[0:1]93        return cost, grads94    if return_inner_v in ['all', True]:95        return cost, updated_ivs96    elif return_inner_v == 'first':97        return cost, updated_ivs[0:1]98    return cost99def _train_tasks_with_for_loop(batch, inner_sub_batch_fn, outer_sub_batch_fn, inner_cost_fn, outer_cost_fn, variables,100                               inner_grad_steps, inner_learning_rate, inner_optimization_step, order,101                               average_across_steps, inner_v, keep_innver_v, outer_v, keep_outer_v, return_inner_v,102                               num_tasks, stop_gradients):103    total_cost = 0104    updated_ivs_to_return = list()105    all_grads = list()106    if isinstance(inner_v, (list, tuple)) and isinstance(inner_v[0], (list, tuple, dict, type(None))):107        inner_v_seq = True108    else:109        inner_v_seq = False110    if isinstance(outer_v, (list, tuple)) and isinstance(outer_v[0], (list, tuple, dict, type(None))):111        outer_v_seq = True112    else:113        outer_v_seq = False114    for i, sub_batch in enumerate(batch.unstack(0, True, num_tasks)):115        if inner_sub_batch_fn is not None:116            inner_sub_batch = inner_sub_batch_fn(sub_batch)117        else:118            inner_sub_batch = sub_batch119        if outer_sub_batch_fn is not None:120            outer_sub_batch = outer_sub_batch_fn(sub_batch)121        else:122            outer_sub_batch = sub_batch123        iv = inner_v[i] if inner_v_seq else inner_v124        ov = outer_v[i] if outer_v_seq else outer_v125        cost, updated_iv, grads = _train_task(inner_sub_batch, outer_sub_batch, inner_cost_fn, outer_cost_fn, variables,126                                              inner_grad_steps, inner_learning_rate, inner_optimization_step, order,127                                              average_across_steps, iv, keep_innver_v, ov, keep_outer_v, False,128                                              num_tasks, stop_gradients)129        if (return_inner_v == 'first' and i == 0) or return_inner_v in ['all', True]:130            updated_ivs_to_return.append(updated_iv)131        total_cost = total_cost + cost132        all_grads.append(grads)133    if order == 1:134        if return_inner_v:135            return total_cost / num_tasks, sum(all_grads) / num_tasks, ivy.Container.concat(updated_ivs_to_return, 0)136        return total_cost / num_tasks, sum(all_grads) / num_tasks137    if return_inner_v:138        return total_cost / num_tasks, ivy.Container.concat(updated_ivs_to_return, 0)139    return total_cost / num_tasks140def _train_tasks(batch, inner_batch_fn, outer_batch_fn, inner_cost_fn, outer_cost_fn, variables, inner_grad_steps,141                 inner_learning_rate, inner_optimization_step, order, average_across_steps, batched, inner_v,142                 keep_innver_v, outer_v, keep_outer_v, return_inner_v, num_tasks, stop_gradients):143    if batched:144        return _train_tasks_batched(145            batch, inner_batch_fn, outer_batch_fn, inner_cost_fn, outer_cost_fn, variables, inner_grad_steps,146            inner_learning_rate, inner_optimization_step, order, average_across_steps, inner_v, keep_innver_v, outer_v,147            keep_outer_v, return_inner_v, num_tasks, stop_gradients)148    return _train_tasks_with_for_loop(149        batch, inner_batch_fn, outer_batch_fn, inner_cost_fn, outer_cost_fn, variables, inner_grad_steps,150        inner_learning_rate, inner_optimization_step, order, average_across_steps, inner_v, keep_innver_v, outer_v,151        keep_outer_v, return_inner_v, num_tasks, stop_gradients)152# Public #153# -------#154# First Order155def fomaml_step(batch, inner_cost_fn, outer_cost_fn, variables, inner_grad_steps, inner_learning_rate,156                inner_optimization_step=gradient_descent_update, inner_batch_fn=None, outer_batch_fn=None,157                average_across_steps=False, batched=True, inner_v=None, keep_inner_v=True, outer_v=None,158                keep_outer_v=True, return_inner_v=False, num_tasks=None, stop_gradients=True):159    """160    Perform step of first order MAML.161    :param batch: The input batch162    :type batch: ivy.Container163    :param inner_cost_fn: callable for the inner loop cost function, receving task-specific sub-batch,164                            inner vars and outer vars165    :type inner_cost_fn: callable166    :param outer_cost_fn: callable for the outer loop cost function, receving task-specific sub-batch,167                            inner vars and outer vars. If None, the cost from the inner loop will also be168                            optimized in the outer loop.169    :type outer_cost_fn: callable, optional170    :param variables: Variables to be optimized during the meta step171    :type variables: ivy.Container172    :param inner_grad_steps: Number of gradient steps to perform during the inner loop.173    :type inner_grad_steps: int174    :param inner_learning_rate: The learning rate of the inner loop.175    :type inner_learning_rate: float176    :param inner_optimization_step: The function used for the inner loop optimization.177                                    Default is ivy.gradient_descent_update.178    :type inner_optimization_step: callable, optional179    :param inner_batch_fn: Function to apply to the task sub-batch, before passing to the inner_cost_fn.180                               Default is None.181    :type inner_batch_fn: callable, optional182    :param outer_batch_fn: Function to apply to the task sub-batch, before passing to the outer_cost_fn.183                               Default is None.184    :type outer_batch_fn: callable, optional185    :param average_across_steps: Whether to average the inner loop steps for the outer loop update. Default is False.186    :type average_across_steps: bool, optional187    :param batched: Whether to batch along the time dimension, and run the meta steps in batch. Default is True.188    :type batched: bool, optional189    :param inner_v: Nested variable keys to be optimized during the inner loop, with same keys and boolean values.190    :type inner_v: dict str or list, optional191    :param keep_inner_v: If True, the key chains in inner_v will be kept, otherwise they will be removed.192                            Default is True.193    :type keep_inner_v: bool, optional194    :param outer_v: Nested variable keys to be optimized during the inner loop, with same keys and boolean values.195    :type outer_v: dict str or list, optional196    :param keep_outer_v: If True, the key chains in inner_v will be kept, otherwise they will be removed.197                            Default is True.198    :type keep_outer_v: bool, optional199    :param return_inner_v: Either 'first', 'all', or False. 'first' means the variables for the first task inner loop200                           will also be returned. variables for all tasks will be returned with 'all'. Default is False.201    :type return_inner_v: str, optional202    :param num_tasks: Number of unique tasks to inner-loop optimize for the meta step. Determined from batch by default.203    :type num_tasks: int, optional204    :param stop_gradients: Whether to stop the gradients of the cost. Default is True.205    :type stop_gradients: bool, optional206    :return: The cost and the gradients with respect to the outer loop variables.207    """208    if num_tasks is None:209        num_tasks = batch.shape[0]210    rets = _train_tasks(211        batch, inner_batch_fn, outer_batch_fn, inner_cost_fn, outer_cost_fn, variables, inner_grad_steps,212        inner_learning_rate, inner_optimization_step, 1, average_across_steps, batched, inner_v, keep_inner_v, outer_v,213        keep_outer_v, return_inner_v, num_tasks, stop_gradients)214    cost = rets[0]215    if stop_gradients:216        cost = ivy.stop_gradient(cost, preserve_type=False)217    grads = rets[1]218    if return_inner_v:219        return cost, grads, rets[2]220    return cost, grads221def reptile_step(batch, cost_fn, variables, inner_grad_steps, inner_learning_rate,222                 inner_optimization_step=gradient_descent_update, batched=True, return_inner_v=False, num_tasks=None,223                 stop_gradients=True):224    """225    Perform step of Reptile.226    :param batch: The input batch227    :type batch: ivy.Container228    :param cost_fn: callable for the cost function, receivng the task-specific sub-batch and variables229    :type cost_fn: callable230    :param variables: Variables to be optimized231    :type variables: ivy.Container232    :param inner_grad_steps: Number of gradient steps to perform during the inner loop.233    :type inner_grad_steps: int234    :param inner_learning_rate: The learning rate of the inner loop.235    :type inner_learning_rate: float236    :param inner_optimization_step: The function used for the inner loop optimization.237                                    Default is ivy.gradient_descent_update.238    :type inner_optimization_step: callable, optional239    :param batched: Whether to batch along the time dimension, and run the meta steps in batch. Default is True.240    :type batched: bool, optional241    :param return_inner_v: Either 'first', 'all', or False. 'first' means the variables for the first task inner loop242                           will also be returned. variables for all tasks will be returned with 'all'. Default is False.243    :type return_inner_v: str, optional244    :param num_tasks: Number of unique tasks to inner-loop optimize for the meta step. Determined from batch by default.245    :type num_tasks: int, optional246    :param stop_gradients: Whether to stop the gradients of the cost. Default is True.247    :type stop_gradients: bool, optional248    :return: The cost and the gradients with respect to the outer loop variables.249    """250    if num_tasks is None:251        num_tasks = batch.shape[0]252    # noinspection PyTypeChecker253    rets = _train_tasks(254        batch, None, None, cost_fn, None, variables, inner_grad_steps, inner_learning_rate, inner_optimization_step,255        1, True, batched, None, True, None, True, return_inner_v, num_tasks, stop_gradients)256    cost = rets[0]257    if stop_gradients:258        cost = ivy.stop_gradient(cost, preserve_type=False)259    grads = rets[1] / inner_learning_rate260    if return_inner_v:261        return cost, grads, rets[2]262    return cost, grads263# Second Order264def maml_step(batch, inner_cost_fn, outer_cost_fn, variables, inner_grad_steps, inner_learning_rate,265              inner_optimization_step=gradient_descent_update, inner_batch_fn=None, outer_batch_fn=None,266              average_across_steps=False, batched=True, inner_v=None, keep_inner_v=True, outer_v=None,267              keep_outer_v=True, return_inner_v=False, num_tasks=None, stop_gradients=True):268    """269    Perform step of vanilla second order MAML.270    :param batch: The input batch271    :type batch: ivy.Container272    :param inner_cost_fn: callable for the inner loop cost function, receing sub-batch, inner vars and outer vars273    :type inner_cost_fn: callable274    :param outer_cost_fn: callable for the outer loop cost function, receving task-specific sub-batch,275                            inner vars and outer vars. If None, the cost from the inner loop will also be276                            optimized in the outer loop.277    :type outer_cost_fn: callable, optional278    :param variables: Variables to be optimized during the meta step279    :type variables: ivy.Container280    :param inner_grad_steps: Number of gradient steps to perform during the inner loop.281    :type inner_grad_steps: int282    :param inner_learning_rate: The learning rate of the inner loop.283    :type inner_learning_rate: float284    :param inner_optimization_step: The function used for the inner loop optimization.285                                    Default is ivy.gradient_descent_update.286    :type inner_optimization_step: callable, optional287    :param inner_batch_fn: Function to apply to the task sub-batch, before passing to the inner_cost_fn.288                               Default is None.289    :type inner_batch_fn: callable, optional290    :param outer_batch_fn: Function to apply to the task sub-batch, before passing to the outer_cost_fn.291                               Default is None.292    :type outer_batch_fn: callable, optional293    :param average_across_steps: Whether to average the inner loop steps for the outer loop update. Default is False.294    :type average_across_steps: bool, optional295    :param batched: Whether to batch along the time dimension, and run the meta steps in batch. Default is True.296    :type batched: bool, optional297    :param inner_v: Nested variable keys to be optimized during the inner loop, with same keys and boolean values.298    :type inner_v: dict str or list, optional299    :param keep_inner_v: If True, the key chains in inner_v will be kept, otherwise they will be removed.300                            Default is True.301    :type keep_inner_v: bool, optional302    :param outer_v: Nested variable keys to be optimized during the inner loop, with same keys and boolean values.303    :type outer_v: dict str or list, optional304    :param keep_outer_v: If True, the key chains in inner_v will be kept, otherwise they will be removed.305                            Default is True.306    :type keep_outer_v: bool, optional307    :param return_inner_v: Either 'first', 'all', or False. 'first' means the variables for the first task inner loop308                           will also be returned. variables for all tasks will be returned with 'all'. Default is False.309    :type return_inner_v: str, optional310    :param num_tasks: Number of unique tasks to inner-loop optimize for the meta step. Determined from batch by default.311    :type num_tasks: int, optional312    :param stop_gradients: Whether to stop the gradients of the cost. Default is True.313    :type stop_gradients: bool, optional314    :return: The cost and the gradients with respect to the outer loop variables.315    """316    if num_tasks is None:317        num_tasks = batch.shape[0]318    unique_outer = outer_v is not None319    cost, grads, *rets = ivy.execute_with_gradients(lambda v: _train_tasks(320        batch, inner_batch_fn, outer_batch_fn, inner_cost_fn, outer_cost_fn,321        variables.set_at_key_chains(v) if unique_outer else v, inner_grad_steps, inner_learning_rate,322        inner_optimization_step, 2, average_across_steps, batched, inner_v, keep_inner_v, outer_v, keep_outer_v,323        return_inner_v, num_tasks, False),324        variables.at_key_chains(outer_v, ignore_none=True)325        if keep_outer_v else variables.prune_key_chains(outer_v, ignore_none=True))326    if stop_gradients:327        cost = ivy.stop_gradient(cost, preserve_type=False)328    # noinspection PyRedundantParentheses...

ragged_factory_ops.py

Source:ragged_factory_ops.py

1# Copyright 2018 The TensorFlow Authors. All Rights Reserved.2#3# Licensed under the Apache License, Version 2.0 (the "License");4# you may not use this file except in compliance with the License.5# You may obtain a copy of the License at6#7#     http://www.apache.org/licenses/LICENSE-2.08#9# Unless required by applicable law or agreed to in writing, software10# distributed under the License is distributed on an "AS IS" BASIS,11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.12# See the License for the specific language governing permissions and13# limitations under the License.14# ==============================================================================15"""Operations for constructing RaggedTensors."""16from __future__ import absolute_import17from __future__ import division18from __future__ import print_function19import numpy as np20from tensorflow.python.framework import constant_op21from tensorflow.python.framework import dtypes22from tensorflow.python.framework import ops23from tensorflow.python.framework import tensor_shape24from tensorflow.python.ops import array_ops25from tensorflow.python.ops.ragged import ragged_tensor26from tensorflow.python.ops.ragged import ragged_tensor_value27from tensorflow.python.util.tf_export import tf_export28#===============================================================================29# Op to construct a constant RaggedTensor from a nested Python list.30#===============================================================================31@tf_export("ragged.constant")32def constant(pylist, dtype=None, ragged_rank=None, inner_shape=None,33             name=None, row_splits_dtype=dtypes.int64):34  """Constructs a constant RaggedTensor from a nested Python list.35  Example:36  >>> tf.ragged.constant([[1, 2], [3], [4, 5, 6]])37  <tf.RaggedTensor [[1, 2], [3], [4, 5, 6]]>38  All scalar values in `pylist` must have the same nesting depth `K`, and the39  returned `RaggedTensor` will have rank `K`.  If `pylist` contains no scalar40  values, then `K` is one greater than the maximum depth of empty lists in41  `pylist`.  All scalar values in `pylist` must be compatible with `dtype`.42  Args:43    pylist: A nested `list`, `tuple` or `np.ndarray`.  Any nested element that44      is not a `list`, `tuple` or `np.ndarray` must be a scalar value45      compatible with `dtype`.46    dtype: The type of elements for the returned `RaggedTensor`.  If not47      specified, then a default is chosen based on the scalar values in48      `pylist`.49    ragged_rank: An integer specifying the ragged rank of the returned50      `RaggedTensor`.  Must be nonnegative and less than `K`. Defaults to51      `max(0, K - 1)` if `inner_shape` is not specified.  Defaults to `max(0, K52      - 1 - len(inner_shape))` if `inner_shape` is specified.53    inner_shape: A tuple of integers specifying the shape for individual inner54      values in the returned `RaggedTensor`.  Defaults to `()` if `ragged_rank`55      is not specified.  If `ragged_rank` is specified, then a default is chosen56      based on the contents of `pylist`.57    name: A name prefix for the returned tensor (optional).58    row_splits_dtype: data type for the constructed `RaggedTensor`'s row_splits.59      One of `tf.int32` or `tf.int64`.60  Returns:61    A potentially ragged tensor with rank `K` and the specified `ragged_rank`,62    containing the values from `pylist`.63  Raises:64    ValueError: If the scalar values in `pylist` have inconsistent nesting65      depth; or if ragged_rank or inner_shape are incompatible with `pylist`.66  """67  def ragged_factory(values, row_splits):68    row_splits = constant_op.constant(row_splits, dtype=row_splits_dtype)69    return ragged_tensor.RaggedTensor.from_row_splits(values, row_splits,70                                                      validate=False)71  with ops.name_scope(name, "RaggedConstant"):72    return _constant_value(ragged_factory, constant_op.constant, pylist, dtype,73                           ragged_rank, inner_shape)74@tf_export(v1=["ragged.constant_value"])75def constant_value(pylist, dtype=None, ragged_rank=None, inner_shape=None,76                   row_splits_dtype="int64"):77  """Constructs a RaggedTensorValue from a nested Python list.78  Warning: This function returns a `RaggedTensorValue`, not a `RaggedTensor`.79  If you wish to construct a constant `RaggedTensor`, use80  [`ragged.constant(...)`](constant.md) instead.81  Example:82  >>> tf.compat.v1.ragged.constant_value([[1, 2], [3], [4, 5, 6]])83  tf.RaggedTensorValue(values=array([1, 2, 3, 4, 5, 6]),84                       row_splits=array([0, 2, 3, 6]))85  All scalar values in `pylist` must have the same nesting depth `K`, and the86  returned `RaggedTensorValue` will have rank `K`.  If `pylist` contains no87  scalar values, then `K` is one greater than the maximum depth of empty lists88  in `pylist`.  All scalar values in `pylist` must be compatible with `dtype`.89  Args:90    pylist: A nested `list`, `tuple` or `np.ndarray`.  Any nested element that91      is not a `list` or `tuple` must be a scalar value compatible with `dtype`.92    dtype: `numpy.dtype`.  The type of elements for the returned `RaggedTensor`.93      If not specified, then a default is chosen based on the scalar values in94      `pylist`.95    ragged_rank: An integer specifying the ragged rank of the returned96      `RaggedTensorValue`.  Must be nonnegative and less than `K`. Defaults to97      `max(0, K - 1)` if `inner_shape` is not specified.  Defaults to `max(0, K98      - 1 - len(inner_shape))` if `inner_shape` is specified.99    inner_shape: A tuple of integers specifying the shape for individual inner100      values in the returned `RaggedTensorValue`.  Defaults to `()` if101      `ragged_rank` is not specified.  If `ragged_rank` is specified, then a102      default is chosen based on the contents of `pylist`.103    row_splits_dtype: data type for the constructed `RaggedTensorValue`'s104      row_splits.  One of `numpy.int32` or `numpy.int64`.105  Returns:106    A `tf.RaggedTensorValue` or `numpy.array` with rank `K` and the specified107    `ragged_rank`, containing the values from `pylist`.108  Raises:109    ValueError: If the scalar values in `pylist` have inconsistent nesting110      depth; or if ragged_rank or inner_shape are incompatible with `pylist`.111  """112  if dtype is not None and isinstance(dtype, dtypes.DType):113    dtype = dtype.as_numpy_dtype114  row_splits_dtype = dtypes.as_dtype(row_splits_dtype).as_numpy_dtype115  def _ragged_factory(values, row_splits):116    row_splits = np.array(row_splits, dtype=row_splits_dtype)117    return ragged_tensor_value.RaggedTensorValue(values, row_splits)118  def _inner_factory(pylist, dtype, shape, name=None):  # pylint: disable=unused-argument119    return np.reshape(np.array(pylist, dtype=dtype), shape)120  return _constant_value(_ragged_factory, _inner_factory, pylist, dtype,121                         ragged_rank, inner_shape)122def _constant_value(ragged_factory, inner_factory, pylist, dtype, ragged_rank,123                    inner_shape):124  """Constructs a constant RaggedTensor or RaggedTensorValue.125  Args:126    ragged_factory: A factory function with the signature:127      `ragged_factory(values, row_splits)`128    inner_factory: A factory function with the signature: `inner_factory(pylist,129      dtype, shape, name)`130    pylist: A nested `list`, `tuple` or `np.ndarray`.131    dtype: Data type for returned value.132    ragged_rank: Ragged rank for returned value.133    inner_shape: Inner value shape for returned value.134  Returns:135    A value returned by `ragged_factory` or `inner_factory`.136  Raises:137    ValueError: If the scalar values in `pylist` have inconsistent nesting138      depth; or if ragged_rank or inner_shape are incompatible with `pylist`.139  """140  if ragged_tensor.is_ragged(pylist):141    raise TypeError("pylist may not be a RaggedTensor or RaggedTensorValue.")142  # np.ndim builds an array, so we short-circuit lists and tuples.143  if not isinstance(pylist, (list, tuple)) and np.ndim(pylist) == 0:144    # Scalar value145    if ragged_rank is not None and ragged_rank != 0:146      raise ValueError("Invalid pylist=%r: incompatible with ragged_rank=%d" %147                       (pylist, ragged_rank))148    if inner_shape is not None and inner_shape:149      raise ValueError(150          "Invalid pylist=%r: incompatible with dim(inner_shape)=%d" %151          (pylist, len(inner_shape)))152    return inner_factory(pylist, dtype, ())153  if ragged_rank is not None and ragged_rank < 0:154    raise ValueError(155        "Invalid ragged_rank=%r: must be nonnegative" % ragged_rank)156  # Find the depth of scalar values in `pylist`.157  scalar_depth, max_depth = _find_scalar_and_max_depth(pylist)158  if scalar_depth is not None:159    if max_depth > scalar_depth:160      raise ValueError("Invalid pylist=%r: empty list nesting is greater "161                       "than scalar value nesting" % pylist)162  # If both inner_shape and ragged_rank were specified, then check that163  # they are compatible with pylist.164  if inner_shape is not None and ragged_rank is not None:165    expected_depth = ragged_rank + len(inner_shape) + 1166    if ((scalar_depth is not None and expected_depth != scalar_depth) or167        (scalar_depth is None and expected_depth < max_depth)):168      raise ValueError(169          "Invalid pylist=%r: incompatible with ragged_rank=%d "170          "and dim(inner_shape)=%d" % (pylist, ragged_rank, len(inner_shape)))171  # Check if the result is a `Tensor`.172  if (ragged_rank == 0 or173      (ragged_rank is None and174       ((max_depth < 2) or175        (inner_shape is not None and max_depth - len(inner_shape) < 2)))):176    return inner_factory(pylist, dtype, inner_shape)177  # Compute default value for inner_shape.178  if inner_shape is None:179    if ragged_rank is None:180      inner_shape = ()181    else:182      inner_shape = _default_inner_shape_for_pylist(pylist, ragged_rank)183  # Compute default value for ragged_rank.184  if ragged_rank is None:185    if scalar_depth is None:186      ragged_rank = max(1, max_depth - 1)187    else:188      ragged_rank = max(1, scalar_depth - 1 - len(inner_shape))189  # Build the splits for each ragged rank, and concatenate the inner values190  # into a single list.191  nested_splits = []192  values = pylist193  for dim in range(ragged_rank):194    nested_splits.append([0])195    concatenated_values = []196    for row in values:197      nested_splits[dim].append(nested_splits[dim][-1] + len(row))198      concatenated_values.extend(row)199    values = concatenated_values200  values = inner_factory(201      values, dtype=dtype, shape=(len(values),) + inner_shape, name="values")202  for row_splits in reversed(nested_splits):203    values = ragged_factory(values, row_splits)204  return values205def _find_scalar_and_max_depth(pylist):206  """Finds nesting depth of scalar values in pylist.207  Args:208    pylist: A nested python `list` or `tuple`.209  Returns:210    A tuple `(scalar_depth, max_depth)`.  `scalar_depth` is the nesting211    depth of scalar values in `pylist`, or `None` if `pylist` contains no212    scalars.  `max_depth` is the maximum depth of `pylist` (including213    empty lists).214  Raises:215    ValueError: If pylist has inconsistent nesting depths for scalars.216  """217  # Check if pylist is not scalar. np.ndim builds an array, so we218  # short-circuit lists and tuples.219  if isinstance(pylist, (list, tuple)) or np.ndim(pylist) != 0:220    scalar_depth = None221    max_depth = 1222    for child in pylist:223      child_scalar_depth, child_max_depth = _find_scalar_and_max_depth(child)224      if child_scalar_depth is not None:225        if scalar_depth is not None and scalar_depth != child_scalar_depth + 1:226          raise ValueError("all scalar values must have the same nesting depth")227        scalar_depth = child_scalar_depth + 1228      max_depth = max(max_depth, child_max_depth + 1)229    return (scalar_depth, max_depth)230  return (0, 0)231def _default_inner_shape_for_pylist(pylist, ragged_rank):232  """Computes a default inner shape for the given python list."""233  def get_inner_shape(item):234    """Returns the inner shape for a python list `item`."""235    if not isinstance(item, (list, tuple)) and np.ndim(item) == 0:236      return ()237    elif item:238      return (len(item),) + get_inner_shape(item[0])239    return (0,)240  def check_inner_shape(item, shape):241    """Checks that `item` has a consistent shape matching `shape`."""242    is_nested = isinstance(item, (list, tuple)) or np.ndim(item) != 0243    if is_nested != bool(shape):244      raise ValueError("inner values have inconsistent shape")245    if is_nested:246      if shape[0] != len(item):247        raise ValueError("inner values have inconsistent shape")248      for child in item:249        check_inner_shape(child, shape[1:])250  # Collapse the ragged layers to get the list of inner values.251  flat_values = pylist252  for dim in range(ragged_rank):253    if not all(254        isinstance(v, (list, tuple)) or np.ndim(v) != 0 for v in flat_values):255      raise ValueError("pylist has scalar values depth %d, but ragged_rank=%d "256                       "requires scalar value depth greater than %d" %257                       (dim + 1, ragged_rank, ragged_rank))258    flat_values = sum((list(v) for v in flat_values), [])259  # Compute the inner shape looking only at the leftmost elements; and then260  # use check_inner_shape to verify that other elements have the same shape.261  inner_shape = get_inner_shape(flat_values)262  check_inner_shape(flat_values, inner_shape)263  return inner_shape[1:]264@tf_export(v1=["ragged.placeholder"])265def placeholder(dtype, ragged_rank, value_shape=None, name=None):266  """Creates a placeholder for a `tf.RaggedTensor` that will always be fed.267  **Important**: This ragged tensor will produce an error if evaluated.268  Its value must be fed using the `feed_dict` optional argument to269  `Session.run()`, `Tensor.eval()`, or `Operation.run()`.270  @compatibility{eager} Placeholders are not compatible with eager execution.271  Args:272    dtype: The data type for the `RaggedTensor`.273    ragged_rank: The ragged rank for the `RaggedTensor`274    value_shape: The shape for individual flat values in the `RaggedTensor`.275    name: A name for the operation (optional).276  Returns:277    A `RaggedTensor` that may be used as a handle for feeding a value, but278    not evaluated directly.279  Raises:280    RuntimeError: if eager execution is enabled281  """282  if ragged_rank == 0:283    return array_ops.placeholder(dtype, value_shape, name)284  with ops.name_scope(name, "RaggedPlaceholder", []):285    flat_shape = tensor_shape.TensorShape([None]).concatenate(value_shape)286    result = array_ops.placeholder(dtype, flat_shape, "flat_values")287    for i in reversed(range(ragged_rank)):288      row_splits = array_ops.placeholder(dtypes.int64, [None],289                                         "row_splits_%d" % i)290      result = ragged_tensor.RaggedTensor.from_row_splits(result, row_splits)...

test_closure.py

Source:test_closure.py

...67        self.run_jit_inner_function(nopython=True)68    @testing.allow_interpreter_mode69    def test_return_closure(self):70        def outer(x):71            def inner():72                return x + 173            return inner74        cfunc = jit(outer)75        self.assertEqual(cfunc(10)(), outer(10)())76class TestInlinedClosure(TestCase):77    """78    Tests for (partial) closure support in njit. The support is partial79    because it only works for closures that can be successfully inlined80    at compile time.81    """82    @tag('important')83    def test_inner_function(self):84        def outer(x):85            def inner(x):86                return x * x87            return inner(x) + inner(x)88        cfunc = njit(outer)89        self.assertEqual(cfunc(10), outer(10))90    @tag('important')91    def test_inner_function_with_closure(self):92        def outer(x):93            y = x + 194            def inner(x):95                return x * x + y96            return inner(x) + inner(x)97        cfunc = njit(outer)98        self.assertEqual(cfunc(10), outer(10))99    @tag('important')100    def test_inner_function_with_closure_2(self):101        def outer(x):102            y = x + 1103            def inner(x):104                return x * y105            y = inner(x)106            return y + inner(x)107        cfunc = njit(outer)108        self.assertEqual(cfunc(10), outer(10))109    @unittest.skipIf(utils.PYVERSION < (3, 0), "needs Python 3")110    def test_inner_function_with_closure_3(self):111        code = """112            def outer(x):113                y = x + 1114                z = 0115                def inner(x):116                    nonlocal z117                    z += x * x118                    return z + y119                return inner(x) + inner(x) + z120        """121        ns = {}122        exec(code.strip(), ns)123        cfunc = njit(ns['outer'])124        self.assertEqual(cfunc(10), ns['outer'](10))125    @tag('important')126    def test_inner_function_nested(self):127        def outer(x):128            def inner(y):129                def innermost(z):130                    return x + y + z131                s = 0132                for i in range(y):133                    s += innermost(i)134                return s135            return inner(x * x)136        cfunc = njit(outer)137        self.assertEqual(cfunc(10), outer(10))138    @tag('important')139    def test_bulk_use_cases(self):140        """ Tests the large number of use cases defined below """141        # jitted function used in some tests142        @njit143        def fib3(n):144            if n < 2:145                return n146            return fib3(n - 1) + fib3(n - 2)147        def outer1(x):148            """ Test calling recursive function from inner """149            def inner(x):150                return fib3(x)151            return inner(x)152        def outer2(x):153            """ Test calling recursive function from closure """154            z = x + 1155            def inner(x):156                return x + fib3(z)157            return inner(x)158        def outer3(x):159            """ Test recursive inner """160            def inner(x):161                if x + y < 2:162                    return 10163                else:164                    inner(x - 1)165            return inner(x)166        def outer4(x):167            """ Test recursive closure """168            y = x + 1169            def inner(x):170                if x + y < 2:171                    return 10172                else:173                    inner(x - 1)174            return inner(x)175        def outer5(x):176            """ Test nested closure """177            y = x + 1178            def inner1(x):179                z = y + x + 2180                def inner2(x):181                    return x + z182                return inner2(x) + y183            return inner1(x)184        def outer6(x):185            """ Test closure with list comprehension in body """186            y = x + 1187            def inner1(x):188                z = y + x + 2189                return [t for t in range(z)]190            return inner1(x)191        _OUTER_SCOPE_VAR = 9192        def outer7(x):193            """ Test use of outer scope var, no closure """194            z = x + 1195            return x + z + _OUTER_SCOPE_VAR196        _OUTER_SCOPE_VAR = 9197        def outer8(x):198            """ Test use of outer scope var, with closure """199            z = x + 1200            def inner(x):201                return x + z + _OUTER_SCOPE_VAR202            return inner(x)203        def outer9(x):204            """ Test closure assignment"""205            z = x + 1206            def inner(x):207                return x + z208            f = inner209            return f(x)210        def outer10(x):211            """ Test two inner, one calls other """212            z = x + 1213            def inner(x):214                return x + z215            def inner2(x):216                return inner(x)217            return inner2(x)218        def outer11(x):219            """ return the closure """220            z = x + 1221            def inner(x):222                return x + z223            return inner224        def outer12(x):225            """ closure with kwarg"""226            z = x + 1227            def inner(x, kw=7):228                return x + z + kw229            return inner(x)230        def outer13(x, kw=7):231            """ outer with kwarg no closure"""232            z = x + 1 + kw233            return z234        def outer14(x, kw=7):235            """ outer with kwarg used in closure"""236            z = x + 1237            def inner(x):238                return x + z + kw239            return inner(x)240        def outer15(x, kw=7):241            """ outer with kwarg as arg to closure"""242            z = x + 1243            def inner(x, kw):244                return x + z + kw245            return inner(x, kw)246        def outer16(x):247            """ closure is generator, consumed locally """248            z = x + 1249            def inner(x):250                yield x + z251            return list(inner(x))252        def outer17(x):253            """ closure is generator, returned """254            z = x + 1255            def inner(x):256                yield x + z257            return inner(x)258        def outer18(x):259            """ closure is generator, consumed in loop """260            z = x + 1261            def inner(x):262                yield x + z263            for i in inner(x):264                t = i265            return t266        def outer19(x):267            """ closure as arg to another closure """268            z1 = x + 1269            z2 = x + 2270            def inner(x):271                return x + z1272            def inner2(f, x):273                return f(x) + z2274            return inner2(inner, x)275        def outer20(x):276            #""" Test calling numpy in closure """277            z = x + 1278            def inner(x):279                return x + numpy.cos(z)280            return inner(x)281        def outer21(x):282            #""" Test calling numpy import as in closure """283            z = x + 1284            def inner(x):285                return x + np.cos(z)286            return inner(x)287        # functions to test that are expected to pass288        f = [outer1, outer2, outer5, outer6, outer7, outer8,289             outer9, outer10, outer12, outer13, outer14,290             outer15, outer19, outer20, outer21]291        for ref in f:292            cfunc = njit(ref)293            var = 10294            self.assertEqual(cfunc(var), ref(var))295        # test functions that are expected to fail296        with self.assertRaises(NotImplementedError) as raises:297            cfunc = jit(nopython=True)(outer3)298            cfunc(var)299        msg = "Unsupported use of op_LOAD_CLOSURE encountered"300        self.assertIn(msg, str(raises.exception))...

Project2_Classification.py

Source:Project2_Classification.py

1# -*- coding: utf-8 -*-2"""3Created on Fri Nov 12 19:11:45 202145@author: Usuario6"""78from matplotlib.pyplot import figure, plot, xlabel, ylabel, legend, show9import sklearn.linear_model as lm10import numpy as np11import pandas as pd12from matplotlib.pylab import (figure, semilogx, loglog, xlabel, ylabel, legend, 13                           title, subplot, show, grid)14import numpy as np15from scipy.io import loadmat16from scipy import stats17import sklearn.linear_model as lm18from sklearn import model_selection19from toolbox_02450 import rlr_validate20import torch21from toolbox_02450 import train_neural_net, draw_neural_net, visualize_decision_boundary22from sklearn.preprocessing import StandardScaler23from sklearn.linear_model import LogisticRegression24from sklearn.model_selection import train_test_split25import statistics as st26import array 2728# Dataset declaration and data cleasing2930df=pd.read_csv('heart.csv')31df.shape32df.info()33df.isnull().sum()34df.info()3536# Creation of matrix X3738X = df.values3940# Outlier removal4142z_scores = stats.zscore(df)43abs_z_scores = np.abs(z_scores)44filtered_entries = (abs_z_scores < 3).all(axis=1)45new_df = df[filtered_entries]4647X_new=new_df48y = X_new['target']49X_new=X_new.drop('target',axis=1)5051#---------Getting dummies for the categorical features--------52X = pd.get_dummies(X_new,columns=['cp','restecg','slope','ca','thal'],drop_first=False)535455X = (X - X.mean())/X.std()5657attributeNames = np.asarray(X.columns)5859N, M = np.shape(X)6061X = np.asarray(X)62y = np.asarray(y)6364#CLASSIFICATION-------------------------6566# Cross Validation Parameters for inner and outer fold.67K_Outer = 568CV_Outer = model_selection.KFold(K_Outer, shuffle=True)6970K_Inner = 571CV_Inner = model_selection.KFold(K_Inner, shuffle=True)7273#Neural Network parameters74h = [1, 3, 5, 7, 9]75max_iter = 100007677#--MODEL ERRORS78Error_test_LR=np.empty((K_Inner,1))79opt_lambda_idx=np.empty((K_Inner,1))80opt_lambda=np.empty((K_Inner,1))8182Error_train_bl_in = np.empty((K_Inner, 1))83Error_test_bl_in = np.empty((K_Inner, 1))84Error_train_bl_out = np.empty((K_Outer, 1))85Error_test_bl_out = np.empty((K_Outer, 1))8687Error_ANN_h = np.empty((K_Inner, 1))88error_in = []89error_out = []90Best_h = np.empty((K_Outer, 1))91Min_Error_h = np.empty((K_Inner, 1 ))92Error_ANN_out = []93949596## ----OUTER CROSS VALIDATION FOLD9798k_out=099for train_index, test_index in CV_Outer.split(X,y):100    print('Outer cross validation fold {0}/{1}:'.format(k_out+1,K_Outer))101    102    # Extract training and test set for the outer cross validation fold103    X_train_outer = X[train_index]104    y_train_outer = y[train_index]105    X_test_outer = X[test_index]106    y_test_outer = y[test_index]107108    # Fit regularized logistic regression model to training data to predict 109110    lambda_interval = np.logspace(-8, 2, 50)111    optim_lambdas = np.empty(K_Outer)112    train_error_rate = np.zeros(len(lambda_interval))113    test_error_rate = np.zeros(len(lambda_interval))114    coefficient_norm = np.zeros(len(lambda_interval))115    116    117    ## -----INNER CROSS VALIDATION FOLD118    119    k_in=0120    for train_index2, test_index2 in CV_Inner.split(X_train_outer,y_train_outer):121        h = [1, 3, 5, 7, 9]122        print('Inner cross validation fold {0}/{1}:'.format(k_in+1,K_Inner))123        124        # Extract inner training and test set for current CV fold125        X_train_inner, X_test_inner, y_train_inner, y_test_inner = train_test_split(X_train_outer, y_train_outer, test_size=.80)126        127        128        #----BASELINE MODEL129        Error_train_bl_in[k_in] = np.sum(y_train_inner != np.argmax(np.bincount(y_train_inner)))/len(y_train_inner)130        Error_test_bl_in[k_in] = np.sum(y_test_inner != np.argmax(np.bincount(y_test_inner)))/len(y_test_inner)131        132        133        134        #vector = np.vectorize(np.int)135        #vector(y_test_inner.numpy())  136        137        138        #----LOGISTIC REGRESSION CLASSIFICATION139        140        # Selection of the best lambda for the inner cross validation fold141        for k in range(0, len(lambda_interval)):142            143            #Creation of the Logistic Regression Model144            mdl = LogisticRegression(penalty='l2', C=1/lambda_interval[k] )145        146            # Training of the model with the inner partition of the CV147            mdl.fit(X_train_inner, y_train_inner)148    149            # Prediction of the model on the inner test partitions150            y_train_est = mdl.predict(X_train_inner).T151            y_test_est = mdl.predict(X_test_inner).T #y_predict152            153            154            # Compute the model erro for each lambda155            train_error_rate[k] = np.sum(y_train_est != y_train_inner) / len(y_train_inner)156            test_error_rate[k] = np.sum(y_test_est != y_test_inner) / len(y_test_inner)157    158            w_est = mdl.coef_[0] 159            coefficient_norm[k] = np.sqrt(np.sum(w_est**2))160        161        #----ARTIFICIAL NEURAL NETWORK FOR CLASSIFICATION162        X_train_inner = torch.Tensor(X_train_outer[train_index2,:] )163        y_train_inner = torch.Tensor(y_train_outer[train_index2] )164        X_test_inner = torch.Tensor(X_train_outer[test_index2,:] )165        y_test_inner = torch.Tensor(y_train_outer[test_index2] )166        167        y_train_inner = y_train_inner.unsqueeze(1)168        error_in = []169            170        for i, j in enumerate(h):  171            172            # Create a model for each h173            inner_ann = lambda: torch.nn.Sequential(174                            torch.nn.Linear(M, h[i]), #M features to H hiden units175                            # 1st transfer function, either Tanh or ReLU:176                            torch.nn.Tanh(),                            177                            torch.nn.Linear(h[i], 1), # H hidden units to 1 output neuron178                            torch.nn.Sigmoid() #Final transfer function179                            )180            loss_fn = torch.nn.BCELoss()181            print('\nTesting h: {0}'.format(j))    182            183            184            # Train the new model185            net, final_loss_in, learning_curve = train_neural_net(inner_ann,186                                                               loss_fn,187                                                               X=X_train_inner,188                                                               y=y_train_inner,189                                                               n_replicates=1,190                                                               max_iter=max_iter)191        192            print('\n\tBest loss: {}\n'.format(final_loss_in))193            194            # Determine estimated class labels for test set195            y_sigmoid_in = net(X_test_inner) # activation of final note, i.e. prediction of network196            y_test_est_in = (y_sigmoid_in > .5).type(dtype=torch.uint8) # threshold output of sigmoidal function197            y_test_in = y_test_inner.type(dtype=torch.uint8)198            # Determine errors and error rate199            e_in = (y_test_est_in != y_test_in)200            error_rate_in = (sum(e_in).type(torch.float)/len(y_test_inner)).data.numpy()201            error_in.append(error_rate_in) # store error rate for current CV fold202            Error_ANN_h[i] = round(np.mean(error_in),4)203            # Determine errors and error rate204            #InnerErrors_h[i] = final_loss_in/y_test_inner.shape[0]205            if (Error_ANN_h[i] < Error_ANN_h[i-1]):206                Besth = j207            else:208                Besth = h[0]209            210        211        212        #Choose the minimum error for given h213        Min_Error_h[k_in] = min(Error_ANN_h)214        215        # Best h for each inner fold        216        Best_h[k_out] = Besth217        218        219        k_in+=1220    221    # COMPUTE THE ERRORS OF THE BEST MODEL FOR THE OUTER FOLD222    223    # Baseline Model224    Error_train_bl_out[k_out] = min(Error_train_bl_in)225    Error_test_bl_out[k_out] = min(Error_test_bl_in)226    227    p=range(len(y_test_outer))228    y_predict_bl=array.array('i',[])229    for i in p:230        y_predict_bl.append(np.argmax(np.bincount(y_test_outer)))231    len(y_predict_bl)232    233    # Logistic Regression234    Error_test_LR[k_out] = np.min(test_error_rate)235    opt_lambda_idx[k_out] = np.argmin(test_error_rate)236    opt_lambda[k_out] = lambda_interval[int(opt_lambda_idx[k_out])]237    238    239    240    LR= LogisticRegression(penalty='l2' , C=1/opt_lambda[k_out].item() )241242    LR.fit(X_train_outer, y_train_outer)243    244    y_predict_LR = LR.predict(X_test_outer).T245    246247    248    249    # Neural Network for outer fold250    # - Create Outer ANN model251    outer_ann = lambda: torch.nn.Sequential(252                            torch.nn.Linear(M, int(np.asarray(Best_h[k_out]))), #M features to H hiden units253                            # 1st transfer function, either Tanh or ReLU:254                            torch.nn.Tanh(),                            255                            torch.nn.Linear(int(np.asarray(Best_h[k_out])), 1), # H hidden units to 1 output neuron256                            torch.nn.Sigmoid() #Final transfer function257                            )258    loss_fn = torch.nn.BCELoss()259    260    # - Training data to pytorch261    X_train_out = torch.Tensor(X[train_index,:] )262    y_train_out = torch.Tensor(y[train_index] )263    X_test_out = torch.Tensor(X[test_index,:] )264    y_test_out = torch.Tensor(y[test_index] )265    266    267    # - Train the net with outer data folds268    y_train_out = y_train_out.unsqueeze(1)269    net, final_loss_out, learning_curve = train_neural_net(outer_ann,270                                                               loss_fn,271                                                               X=X_train_out,272                                                               y=y_train_out,273                                                               n_replicates=1,274                                                               max_iter=max_iter)275    276    # - Compute the errors of the ANN277    # -- Determine estimated class labels for test set278    y_sigmoid_out = net(X_test_out) # activation of final note, i.e. prediction of network279    280    y_test_est_out = (y_sigmoid_out > .5).type(dtype=torch.uint8) # threshold output of sigmoidal function281282    y_predict_ANN = np.concatenate(y_test_est_out.numpy())283    284    y_test_out = y_test_out.type(dtype=torch.uint8)285    286    # -- Determine errors and error rate287    e_out = (y_test_est_out != y_test_out)288    error_rate_out = (sum(e_out).type(torch.float)/len(y_test_out)).data.numpy()289    Error_ANN_out.append(error_rate_out) # store error rate for current CV fold290    Error_ANN_out[k_out] = round(np.mean(error_in),4)291    292    293    k_out+=1294295296297298299300301302303304305306307308
...

Automation Testing Tutorials

Learn to execute automation testing from scratch with LambdaTest Learning Hub. Right from setting up the prerequisites to run your first automation test, to following best practices and diving deeper into advanced test scenarios. LambdaTest Learning Hubs compile a list of step-by-step guides to help you be proficient with different test automation frameworks i.e. Selenium, Cypress, TestNG etc.