Model search is a framework that implements AutoML algorithms for model architecture search at scale

Overview

Model Search

header

Model search (MS) is a framework that implements AutoML algorithms for model architecture search at scale. It aims to help researchers speed up their exploration process for finding the right model architecture for their classification problems (i.e., DNNs with different types of layers).

The library enables you to:

  • Run many AutoML algorithms out of the box on your data - including automatically searching for the right model architecture, the right ensemble of models and the best distilled models.

  • Compare many different models that are found during the search.

  • Create you own search space to customize the types of layers in your neural networks.

The technical description of the capabilities of this framework are found in InterSpeech paper.

While this framework can potentially be used for regression problems, the current version supports classification problems only. Let's start by looking at some classic classification problems and see how the framework can automatically find competitive model architectures.

Getting Started

Let us start with the simplest case. You have a csv file where the features are numbers and you would like to run let AutoML find the best model architecture for you.

Below is a code snippet for doing so:

import model_search
from model_search import constants
from model_search import single_trainer
from model_search.data import csv_data

trainer = single_trainer.SingleTrainer(
    data=csv_data.Provider(
        label_index=0,
        logits_dimension=2,
        record_defaults=[0, 0, 0, 0],
        filename="model_search/data/testdata/csv_random_data.csv"),
    spec=constants.DEFAULT_DNN)

trainer.try_models(
    number_models=200,
    train_steps=1000,
    eval_steps=100,
    root_dir="/tmp/run_example",
    batch_size=32,
    experiment_name="example",
    experiment_owner="model_search_user")

The above code will try 200 different models - all binary classification models, as the logits_dimension is 2. The root directory will have a subdirectory of all models, all of which will be already evaluated. You can open the directory with tensorboard and see all the models with the evaluation metrics.

The search will be performed according to the default specification. That can be found in: model_search/configs/dnn_config.pbtxt.

For more details about the fields and if you want to create your own specification, you can look at: model_search/proto/phoenix_spec.proto.

Now, what if you don't have a csv with the features? The next section shows how to run without a csv.

Non-csv data

To run with non-csv data, you will have to implement a class inherited from the abstract class model_search.data.Provider. This enables us to define our own input_fn and hence customize the feature columns and the task (i.e., the number of classes in the classification task).

class Provider(object, metaclass=abc.ABCMeta):
  """A data provider interface.

  The Provider abstract class that defines three function for Estimator related
  training that return the following:
    * An input function for training and test input functions that return
      features and label batch tensors. It is responsible for parsing the
      dataset and buffering data.
    * The feature_columns for this dataset.
    * problem statement.
  """

  def get_input_fn(self, hparams, mode, batch_size: int):
    """Returns an `input_fn` for train and evaluation.

    Args:
      hparams: tf.HParams for the experiment.
      mode: Defines whether this is training or evaluation. See
        `estimator.ModeKeys`.
      batch_size: the batch size for training and eval.

    Returns:
      Returns an `input_fn` for train or evaluation.
    """

  def get_serving_input_fn(self, hparams):
    """Returns an `input_fn` for serving in an exported SavedModel.

    Args:
      hparams: tf.HParams for the experiment.

    Returns:
      Returns an `input_fn` that takes no arguments and returns a
        `ServingInputReceiver`.
    """

  @abc.abstractmethod
  def number_of_classes(self) -> int:
    """Returns the number of classes. Logits dim for regression."""

  def get_feature_columns(
      self
  ) -> List[Union[feature_column._FeatureColumn,
                  feature_column_v2.FeatureColumn]]:
    """Returns a `List` of feature columns."""

An example of an implementation can be found in model_search/data/csv_data.py.

Once you have this class, you can pass it to model_search.single_trainer.SingleTrainer and your single trainer can now read your data.

Adding your models and architectures to a search space

You can use our platform to test your own existing models.

Our system searches over what we call blocks. We have created an abstract API for an object that resembles a layer in a DNN. All that needs to be implemented for this class is two functions:

class Block(object, metaclass=abc.ABCMeta):
  """Block api for creating a new block."""

  @abc.abstractmethod
  def build(self, input_tensors, is_training, lengths=None):
    """Builds a block for phoenix.

    Args:
      input_tensors: A list of input tensors.
      is_training: Whether we are training. Used for regularization.
      lengths: The lengths of the input sequences in the batch.

    Returns:
      output_tensors: A list of the output tensors.
    """

  @abc.abstractproperty
  def is_input_order_important(self):
    """Is the order of the entries in the input tensor important.

    Returns:
      A bool specifying if the order of the entries in the input is important.
      Examples where the order is important: Input for a cnn layer.
      (e.g., pixels an image). Examples when the order is not important:
      Input for a dense layer.
    """

Once you have implemented your own blocks (i.e., layers), you need to register them with a decorator. Example:

@register_block(
    lookup_name='AVERAGE_POOL_2X2', init_args={'kernel_size': 2}, enum_id=8)
@register_block(
    lookup_name='AVERAGE_POOL_4X4', init_args={'kernel_size': 4}, enum_id=9)
class AveragePoolBlock(Block):
  """Average Pooling layer."""

  def __init__(self, kernel_size=2):
    self._kernel_size = kernel_size

  def build(self, input_tensors, is_training, lengths=None):

(All code above can be found in model_search/blocks.py). Once registered, you can tell the system to search over these blocks by supplying them in blocks_to_use in PhoenixSpec in model_search/proto/phoenix_spec.proto. Namely, if you look at the default specification for dnn found in model_search/configs/dnn_config.pbtxt, you can change the repeated field blocks_to_use and add you own registered blocks.

Note: Our system stacks blocks one on top of each other to create tower architectures that are then going to be ensembled. You can set the minimal and maximal depth allowed in the config to 1 which will change the system to search over which block perform best for the problem - I.e., your blocks can be now an implementation of full classifiers and the system will choose the best one.

Creating a training stand alone binary without writing a main

Now, let's assume you have the data class, but you don't want to write a main function to run it.

We created a simple way to create a main that will just train a dataset and is configurable via flags.

To create it, you need to follow two steps:

  1. You need to register your data provider.

  2. You need to call a help function to create a build rule.

Example: Suppose you have a provider, then you need to register it via a decorator we define it as follows:

@data.register_provider(lookup_name='csv_data_provider', init_args={})
class Provider(data.Provider):
  """A csv data provider."""

  def __init__(self):

The above code can be found in model_search/data/csv_data_for_binary.py.

Next, once you have such library (data provider defined in a .py file and registered), you can supply this library to a help build function an it will create a binary rule as follows:

model_search_oss_binary(
    name = "csv_data_binary",
    dataset_dep = ":csv_data_for_binary",
)

You can also add a test automatically to test integration of your provider with the system as follows:

model_search_oss_test(
    name = "csv_data_for_binary_test",
    dataset_dep = ":csv_data_for_binary",
    problem_type = "dnn",
    extra_args = [
        "--filename=$${TEST_SRCDIR}/model_search/data/testdata/csv_random_data.csv",
    ],
    test_data = [
        "//model_search/data/testdata:csv_random_data",
    ],
)

The above function will create a runable binary. The snippets are taken from the following file: model_search/data/BUILD. The binary is configurable by the flags in model_search/oss_trainer_lib.py.

Distributed Runs

Our system can run a distributed search - I.e., run many search trainer in parallel.

How does it work?

You need to run your binary on multiple machines. Additionally, you need to make one change to configure the bookkeeping of the search.

On a single machine, the bookkeeping is done via a file. For a distributed system however, we need a database.

In order to point our system to the database, you need to set the flags in the file:

model_search/metadata/ml_metadata_db.py

to point to your database.

Once you have done so, the binaries created from the previous section will connect to this database and an async search will begin.

Cloud AutoML

Want to try higher performance AutoML without writing code? Try: https://cloud.google.com/automl-tables

Owner
Google
Google ❤️ Open Source
Google
Implements Stacked-RNN in numpy and torch with manual forward and backward functions

Recurrent Neural Networks Implements simple recurrent network and a stacked recurrent network in numpy and torch respectively. Both flavours implement

Vishal R 1 Nov 16, 2021
Official PyTorch implementation of Synergies Between Affordance and Geometry: 6-DoF Grasp Detection via Implicit Representations

Synergies Between Affordance and Geometry: 6-DoF Grasp Detection via Implicit Representations Zhenyu Jiang, Yifeng Zhu, Maxwell Svetlik, Kuan Fang, Yu

UT-Austin Robot Perception and Learning Lab 63 Jan 03, 2023
Network Pruning That Matters: A Case Study on Retraining Variants (ICLR 2021)

Network Pruning That Matters: A Case Study on Retraining Variants (ICLR 2021)

Duong H. Le 18 Jun 13, 2022
Channel Pruning for Accelerating Very Deep Neural Networks (ICCV'17)

Channel Pruning for Accelerating Very Deep Neural Networks (ICCV'17)

Yihui He 1k Jan 03, 2023
A nutritional label for food for thought.

Lexiscore As a first effort in tackling the theme of information overload in content consumption, I've been working on the lexiscore: a nutritional la

Paul Bricman 34 Nov 08, 2022
DeepVoxels is an object-specific, persistent 3D feature embedding.

DeepVoxels is an object-specific, persistent 3D feature embedding. It is found by globally optimizing over all available 2D observations of

Vincent Sitzmann 196 Dec 25, 2022
Distilled coarse part of LoFTR adapted for compatibility with TensorRT and embedded divices

Coarse LoFTR TRT Google Colab demo notebook This project provides a deep learning model for the Local Feature Matching for two images that can be used

Kirill 46 Dec 24, 2022
Unsupervised Learning of Video Representations using LSTMs

Unsupervised Learning of Video Representations using LSTMs Code for paper Unsupervised Learning of Video Representations using LSTMs by Nitish Srivast

Elman Mansimov 341 Dec 20, 2022
Awesome-AI-books - Some awesome AI related books and pdfs for learning and downloading

Awesome AI books Some awesome AI related books and pdfs for downloading and learning. Preface This repo only used for learning, do not use in business

luckyzhou 1k Jan 01, 2023
Submission to Twitter's algorithmic bias bounty challenge

Twitter Ethics Challenge: Pixel Perfect Submission to Twitter's algorithmic bias bounty challenge, by Travis Hoppe (@metasemantic). Abstract We build

Travis Hoppe 4 Aug 19, 2022
Code for the paper "Benchmarking and Analyzing Point Cloud Classification under Corruptions"

ModelNet-C Code for the paper "Benchmarking and Analyzing Point Cloud Classification under Corruptions". For the latest updates, see: sites.google.com

Jiawei Ren 45 Dec 28, 2022
Official implementation of NeurIPS'21: Implicit SVD for Graph Representation Learning

isvd Official implementation of NeurIPS'21: Implicit SVD for Graph Representation Learning If you find this code useful, you may cite us as: @inprocee

Sami Abu-El-Haija 16 Jan 08, 2023
PromptDet: Expand Your Detector Vocabulary with Uncurated Images

PromptDet: Expand Your Detector Vocabulary with Uncurated Images Paper Website Introduction The goal of this work is to establish a scalable pipeline

103 Dec 20, 2022
EdiBERT, a generative model for image editing

EdiBERT, a generative model for image editing EdiBERT is a generative model based on a bi-directional transformer, suited for image manipulation. The

16 Dec 07, 2022
Cluttered MNIST Dataset

Cluttered MNIST Dataset A setup script will download MNIST and produce mnist/*.t7 files: luajit download_mnist.lua Example usage: local mnist_clutter

DeepMind 50 Jul 12, 2022
A Re-implementation of the paper "A Deep Learning Framework for Character Motion Synthesis and Editing"

What is This This is a simple re-implementation of the paper "A Deep Learning Framework for Character Motion Synthesis and Editing"(1). Only Sections

102 Dec 14, 2022
Recovering Brain Structure Network Using Functional Connectivity

Recovering-Brain-Structure-Network-Using-Functional-Connectivity Framework: Papers: This repository provides a PyTorch implementation of the models ad

5 Nov 30, 2022
A simple editor for captions in .SRT file extension

WaySRT A simple editor for captions in .SRT file extension The program doesn't use any external dependecies, just run: python way_srt.py {file_name.sr

Gustavo Lopes 3 Nov 16, 2022
A tensorflow/keras implementation of StyleGAN to generate images of new Pokemon.

PokeGAN A tensorflow/keras implementation of StyleGAN to generate images of new Pokemon. Dataset The model has been trained on dataset that includes 8

19 Jul 26, 2022
Differentiable Quantum Chemistry (only Differentiable Density Functional Theory and Hartree Fock at the moment)

DQC: Differentiable Quantum Chemistry Differentiable quantum chemistry package. Currently only support differentiable density functional theory (DFT)

75 Dec 02, 2022