|
| 1 | +<!-- Copyright 2017 The TensorFlow Lattice Authors. |
| 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 at |
| 6 | +
|
| 7 | + http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | +
|
| 9 | +Unless required by applicable law or agreed to in writing, software |
| 10 | +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 and |
| 13 | +limitations under the License. |
| 14 | +=============================================================================--> |
| 15 | +# TensorFlow Lattice installation |
| 16 | + |
| 17 | +TensorFlow Lattice runs on Ubuntu and Mac OS X, and requires TensorFlow. |
| 18 | + |
| 19 | +We highly recommend to read [TensorFlow installation |
| 20 | +instructions](https://www.tensorflow.org/install), especially [Installing |
| 21 | +TensorFlow on Ubuntu](https://www.tensorflow.org/install/install_linux) to |
| 22 | +understand virtualenv and pip, and [Installing TensorFlow from |
| 23 | +Sources](https://www.tensorflow.org/install/install_sources). |
| 24 | + |
| 25 | +# Install the prebuilt pip package |
| 26 | + |
| 27 | +## Activate virtualenv |
| 28 | +If using virtualenv, activate your virtualenv for the rest of the installation, |
| 29 | +otherwise skip this step: |
| 30 | + |
| 31 | +``` shell |
| 32 | +~$ virtualenv --system-site-packages tensorflow-lattice # for Python 2.7 |
| 33 | +~$ virtualenv --system-site-packages -p python3 tensorflow-lattice # for Python 3.n |
| 34 | +``` |
| 35 | + |
| 36 | +Here you can change `tensorflow-lattice` to another target directory you want to |
| 37 | +use. |
| 38 | + |
| 39 | +```shell |
| 40 | +~$ source tensorflow-lattice/bin/activate # bash, sh, ksh, or zsh |
| 41 | +~$ source tensorflow-lattice/bin/activate.csh # csh or tcsh |
| 42 | +``` |
| 43 | + |
| 44 | +## Install pip packages. |
| 45 | +You can use pip install to install tensorflow-lattice pip package. |
| 46 | + |
| 47 | +```shell |
| 48 | +(tensorflow-lattice)$ pip install --upgrade tensorflow-lattice # for Python 2.7 |
| 49 | +(tensorflow-lattice)$ pip3 install --upgrade tensorflow-lattice # for Python 3.n |
| 50 | +(tensorflow-lattice)$ pip install --upgrade tensorflow-lattice-gpu # for Python 2.7 and GPU |
| 51 | +(tensorflow-lattice)$ pip3 install --upgrade tensorflow-lattice-gpu # for Python 3.n and GPU |
| 52 | +``` |
| 53 | +Our custom operators do not have GPU kernels. The main difference |
| 54 | +between `tensorflow-lattice-gpu` and `tensorflow-lattice` pip package is that |
| 55 | +the former requires `tensorflow-gpu` pip package whereas the latter requires |
| 56 | +`tensorflow` pip package. |
| 57 | + |
| 58 | +## Test TensorFlow and TensorFlow Lattice |
| 59 | + |
| 60 | +Run the following python script to test TensorFlow Lattice. |
| 61 | + |
| 62 | +```python |
| 63 | +import tensorflow as tf |
| 64 | +import tensorflow_lattice as tfl |
| 65 | + |
| 66 | +x = tf.placeholder(tf.float32, shape=(None, 2)) |
| 67 | +(y, _, _, _) = tfl.lattice_layer(x, lattice_sizes=(2, 2)) |
| 68 | + |
| 69 | +with tf.Session() as sess: |
| 70 | + sess.run(tf.global_variables_initializer()) |
| 71 | + print(sess.run(y, feed_dict={x: [[0.0, 0.0]]})) |
| 72 | +``` |
| 73 | + |
| 74 | +Now you are ready to use *TensorFlow Lattice*, and you can stop here unless you |
| 75 | +want to build TensorFlow Lattice from the source. |
| 76 | + |
| 77 | +# Build TensorFlow Lattice and TensorFlow pip package from the source. |
| 78 | +You can also build TensorFlow Lattice packages from the source. |
| 79 | +For this, you will need to compile all libraries using |
| 80 | +[Bazel](https://bazel.build) against TensorFlow headers. |
| 81 | + |
| 82 | + |
| 83 | +We will show how to build TensorFlow and TensorFlow Lattice pip package using |
| 84 | +Bazel, and install it to your virtualenv. |
| 85 | + |
| 86 | +## Activate virtualenv |
| 87 | + |
| 88 | +If using virtualenv, activate your virtualenv for the rest of the installation, |
| 89 | +otherwise skip this step: |
| 90 | + |
| 91 | +```shell |
| 92 | +~$ source $VIRTUALENV_PATH/bin/activate # bash, sh, ksh, or zsh |
| 93 | +~$ source $VIRTUALENV_PATH/bin/activate.csh # csh or tcsh |
| 94 | +``` |
| 95 | + |
| 96 | +or if you are using virtualenv for the first time, |
| 97 | + |
| 98 | +```shell |
| 99 | +~$ sudo apt-get install python-virtualenv |
| 100 | +~$ virtualenv --system-site-packages tensorflow-lattice |
| 101 | +~$ source ~/tensorflow-lattice/bin/activate # bash, sh, ksh, or zsh |
| 102 | +~$ source ~/tensorflow-lattice/bin/activate.csh # csh or tcsh |
| 103 | +``` |
| 104 | +## Prepare TensorFlow envirnoment for Linux. |
| 105 | + |
| 106 | +Please follow instructions in [Prepare environment for |
| 107 | +Linux](https://www.tensorflow.org/install/install_sources#prepare_environment_for_linux) |
| 108 | +to setup the environment for TensorFlow. |
| 109 | + |
| 110 | +## Clone the TensorFlow Lattice repository. |
| 111 | + |
| 112 | +Let us clone the TensorFlow Lattice repository, which contains TensorFlow as a |
| 113 | +submodule: |
| 114 | + |
| 115 | +```shell |
| 116 | +(tensorflow-lattice)~$ git clone --recursive https://github.com/tensorflow/lattice.git |
| 117 | +``` |
| 118 | + |
| 119 | +## Configure TensorFlow and build TensorFlow pip package. |
| 120 | + |
| 121 | +### Configure TensorFlow |
| 122 | + |
| 123 | +We now need to configure TensorFlow options. See [Configure the |
| 124 | +installation](https://www.tensorflow.org/install/install_sources#configure_the_installation) |
| 125 | +for the details. |
| 126 | + |
| 127 | +```shell |
| 128 | +(tensorflow-lattice)~$ cd lattice |
| 129 | +(tensorflow-lattice)~/lattice$ cd tensorflow |
| 130 | +(tensorflow-lattice)~/lattice/tensorflow$ ./configure |
| 131 | +``` |
| 132 | + |
| 133 | +### Build TensorFlow pip packaging script |
| 134 | + |
| 135 | +We are ready to build the TensorFlow pip package. See [Build the pip |
| 136 | +package](https://www.tensorflow.org/install/install_sources#build_the_pip_package) |
| 137 | +for the details. |
| 138 | + |
| 139 | +To build a pip package for TensorFlow with CPU-only support: |
| 140 | + |
| 141 | +```shell |
| 142 | +(tensorflow-lattice)~/lattice/tensorflow$ bazel build \ |
| 143 | + --config=opt \ |
| 144 | + tensorflow/tools/pip_package:build_pip_package |
| 145 | +``` |
| 146 | + |
| 147 | +To build a pip package for TensorFlow with GPU support: |
| 148 | + |
| 149 | +```shell |
| 150 | +(tensorflow-lattice)~/lattice/tensorflow$ bazel build \ |
| 151 | + --config=cuda \ |
| 152 | + tensorflow/tools/pip_package:build_pip_package |
| 153 | +``` |
| 154 | + |
| 155 | +### Install TensorFlow pip package |
| 156 | + |
| 157 | +```shell |
| 158 | +(tensorflow-lattice)~/lattice/tensorflow$ bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg |
| 159 | +(tensorflow-lattice)~/lattice/tensorflow$ pip install /tmp/tensorflow_pkg/*.whl |
| 160 | +``` |
| 161 | + |
| 162 | +### Build TensorFlow Lattice pip packaging script |
| 163 | + |
| 164 | +To build a pip package for TensorFlow with CPU-only support: |
| 165 | + |
| 166 | +```shell |
| 167 | +(tensorflow-lattice)~/$ cd ~/lattice |
| 168 | +(tensorflow-lattice)~/lattice$ bazel build \ |
| 169 | + --config=opt :pip_pkg |
| 170 | +``` |
| 171 | + |
| 172 | +### Install TensorFlow Lattice pip package |
| 173 | + |
| 174 | +```shell |
| 175 | +(tensorflow-lattice)~/lattice$ bazel-bin/pip_pkg /tmp/tensorflow_lattice_pkg |
| 176 | +(tensorflow-lattice)~/lattice$ pip install /tmp/tensorflow_lattice_pkg/*.whl |
| 177 | +``` |
| 178 | + |
| 179 | +### Test TensorFlow and TensorFlow Lattice |
| 180 | +```shell |
| 181 | +(tensorflow-lattice)~/lattice$ cd examples |
| 182 | +(tensorflow-lattice)~/lattice/examples$ python test.py |
| 183 | +``` |
| 184 | + |
| 185 | +test.py is a simple python script. |
| 186 | + |
| 187 | +```python |
| 188 | +import tensorflow as tf |
| 189 | +import tensorflow_lattice as tfl |
| 190 | + |
| 191 | +x = tf.placeholder(tf.float32, shape=(None, 2)) |
| 192 | +(y, _, _, _) = tfl.lattice_layer(x, lattice_sizes=(2, 2)) |
| 193 | + |
| 194 | +with tf.Session() as sess: |
| 195 | + sess.run(tf.global_variables_initializer()) |
| 196 | + print(sess.run(y, feed_dict={x: [[0.0, 0.0]]})) |
| 197 | +``` |
0 commit comments