bailulu
/
mindspore_yolo
Not watched
1
0
Fork 0
Code
Releases 0 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
Browse Source

ADD file via upload

master
zhangmeng1 3 years ago
parent
a0c177f3f7
commit
532d800178
1 changed files with 660 additions and 0 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +660
    -0
      Project_preparation/preparation_02/tf04_function_and_auto_graph.ipynb

+ 660
- 0
Project_preparation/preparation_02/tf04_function_and_auto_graph.ipynb View File

@@ -0,0 +1,660 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"2.2.0\n",
"sys.version_info(major=3, minor=6, micro=9, releaselevel='final', serial=0)\n",
"matplotlib 3.3.4\n",
"numpy 1.19.5\n",
"pandas 1.1.5\n",
"sklearn 0.24.2\n",
"tensorflow 2.2.0\n",
"tensorflow.keras 2.3.0-tf\n"
]
}
],
"source": [
"import matplotlib as mpl\n",
"import matplotlib.pyplot as plt\n",
"%matplotlib inline\n",
"import numpy as np\n",
"import sklearn\n",
"import pandas as pd\n",
"import os\n",
"import sys\n",
"import time\n",
"import tensorflow as tf\n",
"\n",
"from tensorflow import keras\n",
"\n",
"print(tf.__version__)\n",
"print(sys.version_info)\n",
"for module in mpl, np, pd, sklearn, tf, keras:\n",
" print(module.__name__, module.__version__)"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tf.Tensor(-0.95021296, shape=(), dtype=float32)\n",
"tf.Tensor([-0.95021296 -0.917915 ], shape=(2,), dtype=float32)\n",
"tf.Tensor(-0.95021296, shape=(), dtype=float32)\n",
"tf.Tensor([-0.95021296 -0.917915 ], shape=(2,), dtype=float32)\n",
"True\n",
"<function scaled_elu at 0x7f971c419268>\n",
"<tensorflow.python.eager.def_function.Function object at 0x7f971c408e48>\n"
]
}
],
"source": [
"# tf.function and auto-graph.\n",
"#自己实现一下elu激活函数,如果scale不为1，那就是selu\n",
"def scaled_elu(z, scale=1.0, alpha=1.0):\n",
" # z >= 0 ? scale * z : scale * alpha * tf.nn.elu(z)\n",
" is_positive = tf.greater_equal(z, 0.0)\n",
"# return scale * tf.where(is_positive, z, alpha * tf.nn.elu(z))\n",
" return scale * tf.where(is_positive, z, alpha * (tf.math.exp(z)-1))\n",
"\n",
"#运行一下，这还是py函数\n",
"print(scaled_elu(tf.constant(-3.)))\n",
"print(scaled_elu(tf.constant([-3., -2.5])))\n",
"\n",
"#把py实现的函数变为图实现的函数\n",
"#scaled_elu_tf就是图\n",
"scaled_elu_tf = tf.function(scaled_elu)\n",
"print(scaled_elu_tf(tf.constant(-3.)))\n",
"print(scaled_elu_tf(tf.constant([-3., -2.5])))\n",
"\n",
"#可以通过这种方式找回原来的py函数\n",
"print(scaled_elu_tf.python_function is scaled_elu)\n",
"print(scaled_elu)\n",
"print(scaled_elu_tf)#tf的函数的执行效率比较高"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"1.03 s ± 25.4 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n",
"868 ms ± 6.27 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
]
}
],
"source": [
"#我们来测试一下性能，100万个数\n",
"%timeit scaled_elu(tf.random.normal((10000, 10000)))\n",
"%timeit scaled_elu_tf(tf.random.normal((10000, 10000)))"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"<tensorflow.python.eager.def_function.Function object at 0x7f7f1051d780>\n",
"tf.Tensor(1.9999981, shape=(), dtype=float32)\n"
]
}
],
"source": [
"# 1 + 1/2 + 1/2^2 + ... + 1/2^n\n",
"#加了@tf.function装饰后就变为图结果，但是输入类型上不会有变化\n",
"@tf.function\n",
"def converge_to_2(n_iters):\n",
" total = tf.constant(0.)\n",
" increment = tf.constant(1.)\n",
" for _ in range(n_iters):\n",
" total += increment\n",
" increment /= 2.0\n",
" return total\n",
"print(converge_to_2)\n",
"print(converge_to_2(20))"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
"#如何看tf的图的代码\n",
"def display_tf_code(func):\n",
" code = tf.autograph.to_code(func)\n",
" from IPython.display import display, Markdown\n",
" display(Markdown('```python\\n{}\\n```'.format(code)))"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"data": {
"text/markdown": [
"```python\n",
"def tf__scaled_elu(z, scale=None, alpha=None):\n",
" do_return = False\n",
" retval_ = ag__.UndefinedReturnValue()\n",
" with ag__.FunctionScope('scaled_elu', 'fscope', ag__.ConversionOptions(recursive=True, user_requested=True, optional_features=(), internal_convert_user_code=True)) as fscope:\n",
" is_positive = ag__.converted_call(tf.greater_equal, (z, 0.0), None, fscope)\n",
" try:\n",
" do_return = True\n",
" retval_ = fscope.mark_return_value((scale * ag__.converted_call(tf.where, (is_positive, z, (alpha * (ag__.converted_call(tf.math.exp, (z,), None, fscope) - 1))), None, fscope)))\n",
" except:\n",
" do_return = False\n",
" raise\n",
" (do_return,)\n",
" return ag__.retval(retval_)\n",
"\n",
"```"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"#传普通py函数,返回的是tf图的代码\n",
"display_tf_code(scaled_elu)"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [
{
"data": {
"text/markdown": [
"```python\n",
"def tf__converge_to_2(n_iters):\n",
" do_return = False\n",
" retval_ = ag__.UndefinedReturnValue()\n",
" with ag__.FunctionScope('converge_to_2', 'fscope', ag__.ConversionOptions(recursive=True, user_requested=True, optional_features=(), internal_convert_user_code=True)) as fscope:\n",
" total = ag__.converted_call(tf.constant, (0.0,), None, fscope)\n",
" increment = ag__.converted_call(tf.constant, (1.0,), None, fscope)\n",
"\n",
" def get_state():\n",
" return (total, increment)\n",
"\n",
" def set_state(loop_vars):\n",
" nonlocal total, increment\n",
" (total, increment) = loop_vars\n",
"\n",
" def loop_body(itr):\n",
" nonlocal total, increment\n",
" _ = itr\n",
" total += increment\n",
" increment /= 2.0\n",
" ag__.for_stmt(ag__.converted_call(range, (n_iters,), None, fscope), None, loop_body, get_state, set_state, ('total', 'increment'), {})\n",
" try:\n",
" do_return = True\n",
" retval_ = fscope.mark_return_value(total)\n",
" except:\n",
" do_return = False\n",
" raise\n",
" (do_return,)\n",
" return ag__.retval(retval_)\n",
"\n",
"```"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"#这个的前提是去除converge_to_2的装饰\n",
"# 因为converge_to_2有@tf.function标注，去掉应该就没问题了。to_code函数的输入是module,\n",
"# class, method, function, traceback, frame, or code object。不能是tf function.\n",
"display_tf_code(converge_to_2.python_function)"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tf.Tensor(21.0, shape=(), dtype=float32)\n"
]
}
],
"source": [
"#tf要把变量定义在函数外面，不能放里边\n",
"var = tf.Variable(0.)\n",
"\n",
"@tf.function\n",
"def add_21():\n",
" return var.assign_add(21) # += \n",
"\n",
"print(add_21())"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Python inputs incompatible with input_signature:\n",
" inputs: (\n",
" tf.Tensor([1. 2. 3.], shape=(3,), dtype=float32))\n",
" input_signature: (\n",
" TensorSpec(shape=(None,), dtype=tf.int32, name='x'))\n",
"--------------------------------------------------\n",
"tf.Tensor([ 1 8 27], shape=(3,), dtype=int32)\n",
"<tensorflow.python.eager.def_function.Function object at 0x7f7e3ef13fd0>\n"
]
}
],
"source": [
"#cube计算立方，py是泛型设计，我们通过input_signature加类型限制可以防止调错\n",
"# @tf.function(input_signature=[tf.TensorSpec([None], tf.int32, name='x')])\n",
"@tf.function\n",
"def cube(z):\n",
" return tf.pow(z, 3)\n",
"\n",
"try:\n",
" print(cube(tf.constant([1., 2., 3.])))\n",
"except ValueError as ex:\n",
" print(ex)\n",
"\n",
"print('-'*50)\n",
"#这行没问题\n",
"print(cube(tf.constant([1, 2, 3])))\n",
"print(cube)"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [
{
"ename": "ValueError",
"evalue": "invalid literal for int() with base 10: 'abc'",
"output_type": "error",
"traceback": [
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
"\u001b[0;31mValueError\u001b[0m Traceback (most recent call last)",
"\u001b[0;32m<ipython-input-13-1e11daa61250>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'abc'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
"\u001b[0;31mValueError\u001b[0m: invalid literal for int() with base 10: 'abc'"
]
}
],
"source": [
"int('abc')"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"<tensorflow.python.eager.function.ConcreteFunction object at 0x7f7edb682e48>\n",
"<tensorflow.python.eager.def_function.Function object at 0x7f7e3ef13fd0>\n",
"tf.Tensor([ 1 8 27], shape=(3,), dtype=int32)\n"
]
}
],
"source": [
"# @tf.function py func -> tf graph\n",
"# get_concrete_function -> 给tf.function add input signature -> SavedModel\n",
"\n",
"cube_func_int32 = cube.get_concrete_function(\n",
" tf.TensorSpec([None], tf.int32))\n",
"print(cube_func_int32)\n",
"print(cube)\n",
"\n",
"try:\n",
" print(cube_func_int32(tf.constant([1, 2, 3])))\n",
"except Exception as ex:\n",
" print(ex)"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"<tensorflow.python.eager.function.ConcreteFunction object at 0x7f7edb682e48>\n",
"<tensorflow.python.eager.function.ConcreteFunction object at 0x7f7edb682e48>\n",
"True\n"
]
}
],
"source": [
"#我们只要看原来函数和新生成的是否一致\n",
"# print(cube_func_int32 is cube.get_concrete_function())\n",
"print(cube.get_concrete_function(\n",
" tf.constant([1, 2, 3])))\n",
"print(cube_func_int32)\n",
"print(cube_func_int32 is cube.get_concrete_function(\n",
" tf.constant([1, 2, 3])))"
]
},
{
"cell_type": "code",
"execution_count": 18,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"<tensorflow.python.eager.function.ConcreteFunction object at 0x7f7edb682e48>\n"
]
},
{
"data": {
"text/plain": [
"<tensorflow.python.framework.func_graph.FuncGraph at 0x7f7e3ef7bef0>"
]
},
"execution_count": 18,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"print(cube_func_int32)\n",
"cube_func_int32.graph"
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[<tf.Operation 'x' type=Placeholder>,\n",
" <tf.Operation 'Pow/y' type=Const>,\n",
" <tf.Operation 'Pow' type=Pow>,\n",
" <tf.Operation 'Identity' type=Identity>]"
]
},
"execution_count": 19,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"#看下图定义都有哪些操作,了解即可\n",
"cube_func_int32.graph.get_operations()"
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {
"scrolled": true
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"name: \"x\"\n",
"op: \"Placeholder\"\n",
"attr {\n",
" key: \"_user_specified_name\"\n",
" value {\n",
" s: \"x\"\n",
" }\n",
"}\n",
"attr {\n",
" key: \"dtype\"\n",
" value {\n",
" type: DT_INT32\n",
" }\n",
"}\n",
"attr {\n",
" key: \"shape\"\n",
" value {\n",
" shape {\n",
" dim {\n",
" size: -1\n",
" }\n",
" }\n",
" }\n",
"}\n",
"\n"
]
}
],
"source": [
"pow_op = cube_func_int32.graph.get_operations()[0]\n",
"print(pow_op)"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[]\n",
"--------------------------------------------------\n",
"[<tf.Tensor 'z:0' shape=(None,) dtype=int32>]\n"
]
}
],
"source": [
"print(list(pow_op.inputs))\n",
"print('-'*50)\n",
"print(list(pow_op.outputs))\n"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<tf.Operation 'z' type=Placeholder>"
]
},
"execution_count": 13,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"#Placeholder用来放输入的地方，2.0中不需要，图中依然保留了\n",
"cube_func_int32.graph.get_operation_by_name(\"z\")"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<tf.Tensor 'z:0' shape=(None,) dtype=int32>"
]
},
"execution_count": 16,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"cube_func_int32.graph.get_tensor_by_name(\"z:0\")"
]
},
{
"cell_type": "code",
"execution_count": 34,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"node {\n",
" name: \"z\"\n",
" op: \"Placeholder\"\n",
" attr {\n",
" key: \"_user_specified_name\"\n",
" value {\n",
" s: \"z\"\n",
" }\n",
" }\n",
" attr {\n",
" key: \"dtype\"\n",
" value {\n",
" type: DT_INT32\n",
" }\n",
" }\n",
" attr {\n",
" key: \"shape\"\n",
" value {\n",
" shape {\n",
" dim {\n",
" size: -1\n",
" }\n",
" }\n",
" }\n",
" }\n",
"}\n",
"node {\n",
" name: \"Pow/y\"\n",
" op: \"Const\"\n",
" attr {\n",
" key: \"dtype\"\n",
" value {\n",
" type: DT_INT32\n",
" }\n",
" }\n",
" attr {\n",
" key: \"value\"\n",
" value {\n",
" tensor {\n",
" dtype: DT_INT32\n",
" tensor_shape {\n",
" }\n",
" int_val: 3\n",
" }\n",
" }\n",
" }\n",
"}\n",
"node {\n",
" name: \"Pow\"\n",
" op: \"Pow\"\n",
" input: \"z\"\n",
" input: \"Pow/y\"\n",
" attr {\n",
" key: \"T\"\n",
" value {\n",
" type: DT_INT32\n",
" }\n",
" }\n",
"}\n",
"node {\n",
" name: \"Identity\"\n",
" op: \"Identity\"\n",
" input: \"Pow\"\n",
" attr {\n",
" key: \"T\"\n",
" value {\n",
" type: DT_INT32\n",
" }\n",
" }\n",
"}\n",
"versions {\n",
" producer: 175\n",
"}"
]
},
"execution_count": 34,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"#打印出来看看图信息\n",
"cube_func_int32.graph.as_graph_def()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.6.9"
}
},
"nbformat": 4,
"nbformat_minor": 2
}

Write Preview
Loading…
Cancel
Save