Running Mode

Ascend GPU Model Running

View Source On Gitee

Overview

There are three execution modes: single operator, common function, and network training model.

Note: This document is applicable to GPU and Ascend environments.

Executing a Single Operator

Execute a single operator and output the result.

A code example is as follows:

import numpy as np
import mindspore.nn as nn
from mindspore import context, Tensor

context.set_context(mode=context.GRAPH_MODE, device_target="GPU")

conv = nn.Conv2d(3, 4, 3, bias_init='zeros')
input_data = Tensor(np.ones([1, 3, 5, 5]).astype(np.float32))
output = conv(input_data)
print(output.asnumpy())

The output is as follows:

[[[[ 0.06022915  0.06149777  0.06149777  0.06149777  0.01145121]
   [ 0.06402162  0.05889071  0.05889071  0.05889071 -0.00933781]
   [ 0.06402162  0.05889071  0.05889071  0.05889071 -0.00933781]
   [ 0.06402162  0.05889071  0.05889071  0.05889071 -0.00933781]
   [ 0.02712326  0.02096302  0.02096302  0.02096302 -0.01119636]]

  [[-0.0258286  -0.03362969 -0.03362969 -0.03362969 -0.00799183]
   [-0.0513729  -0.06778982 -0.06778982 -0.06778982 -0.03168458]
   [-0.0513729  -0.06778982 -0.06778982 -0.06778982 -0.03168458]
   [-0.0513729  -0.06778982 -0.06778982 -0.06778982 -0.03168458]
   [-0.04186669 -0.07266843 -0.07266843 -0.07266843 -0.04836193]]

  [[-0.00840744 -0.03043237 -0.03043237 -0.03043237  0.00172079]
   [ 0.00401019 -0.03755453 -0.03755453 -0.03755453 -0.00851137]
   [ 0.00401019 -0.03755453 -0.03755453 -0.03755453 -0.00851137]
   [ 0.00401019 -0.03755453 -0.03755453 -0.03755453 -0.00851137]
   [ 0.00270888 -0.03718876 -0.03718876 -0.03718876 -0.03043662]]

  [[-0.00982172  0.02009856  0.02009856  0.02009856  0.03327979]
   [ 0.02529106  0.04035065  0.04035065  0.04035065  0.01782833]
   [ 0.02529106  0.04035065  0.04035065  0.04035065  0.01782833]
   [ 0.02529106  0.04035065  0.04035065  0.04035065  0.01782833]
   [ 0.01015155  0.00781826  0.00781826  0.00781826 -0.02884173]]]]

Note: Due to random factors in weight initialization, the actual output results may be different, which is for reference only.

Executing a Common Function

Combine multiple operators into a function, execute these operators by calling the function, and output the result.

A code example is as follows:

import numpy as np
from mindspore import context, Tensor
import mindspore.ops as ops

context.set_context(mode=context.GRAPH_MODE, device_target="GPU")

def add_func(x, y):
    z = ops.add(x, y)
    z = ops.add(z, x)
    return z

x = Tensor(np.ones([3, 3], dtype=np.float32))
y = Tensor(np.ones([3, 3], dtype=np.float32))
output = add_func(x, y)
print(output.asnumpy())

The output is as follows:

[[3. 3. 3.]
 [3. 3. 3.]
 [3. 3. 3.]]

Executing a Network Model

The Model API of MindSpore is an advanced API used for training and validation. Layers with the training or inference function can be combined into an object. The training, inference, and prediction functions can be implemented by calling the train, eval, and predict APIs, respectively.

MindSpore does not support the use of multiple threads for training, inference, and prediction functions.

You can transfer the initialized Model APIs such as the network, loss function, and optimizer as required. You can also configure amp_level to implement mixed precision and configure metrics to implement model evaluation.

Executing the network model will generate a kernel_meta directory under the execution directory, and save the operator cache files generated by network compilation to this directory during execution, including .o, .info and .json files. If the user executes the same network model again, or only some changes are made, MindSpore will automatically call the reusable operator cache file in the kernel_meta directory, which significantly reduces network compilation time and improves execution performance. For details, please refer to Incremental Operator Build

Before executing the network, download and unzip the required dataset to the specified directory in jupyter notebook:

mkdir -p ./datasets/MNIST_Data/train ./datasets/MNIST_Data/test
wget -NP ./datasets/MNIST_Data/train https://mindspore-website.obs.myhuaweicloud.com/notebook/datasets/mnist/train-labels-idx1-ubyte --no-check-certificate
wget -NP ./datasets/MNIST_Data/train https://mindspore-website.obs.myhuaweicloud.com/notebook/datasets/mnist/train-images-idx3-ubyte --no-check-certificate
wget -NP ./datasets/MNIST_Data/test https://mindspore-website.obs.myhuaweicloud.com/notebook/datasets/mnist/t10k-labels-idx1-ubyte --no-check-certificate
wget -NP ./datasets/MNIST_Data/test https://mindspore-website.obs.myhuaweicloud.com/notebook/datasets/mnist/t10k-images-idx3-ubyte --no-check-certificate

The directory structure of the downloaded dataset file is as follows:

./datasets/MNIST_Data
├── test
│   ├── t10k-images-idx3-ubyte
│   └── t10k-labels-idx1-ubyte
└── train
    ├── train-images-idx3-ubyte
    └── train-labels-idx1-ubyte

2 directories, 4 files

Executing a Training Model

Call the train API of Model to implement training.

A code example is as follows:

import os
import mindspore.dataset.vision.c_transforms as CV
from mindspore.dataset.vision import Inter
import mindspore.dataset as ds
import mindspore.dataset.transforms.c_transforms as CT
import mindspore.nn as nn
from mindspore import context, Model
from mindspore import dtype as mstype
from mindspore.common.initializer import Normal
from mindspore.train.callback import LossMonitor, ModelCheckpoint, CheckpointConfig


def create_dataset(data_path, batch_size=32, repeat_size=1,
                   num_parallel_workers=1):
    """
    create dataset for train or test
    """
    # define dataset
    mnist_ds = ds.MnistDataset(data_path)

    resize_height, resize_width = 32, 32
    rescale = 1.0 / 255.0
    shift = 0.0
    rescale_nml = 1 / 0.3081
    shift_nml = -1 * 0.1307 / 0.3081

    # define map operations
    resize_op = CV.Resize((resize_height, resize_width), interpolation=Inter.LINEAR)  # Bilinear mode
    rescale_nml_op = CV.Rescale(rescale_nml, shift_nml)
    rescale_op = CV.Rescale(rescale, shift)
    hwc2chw_op = CV.HWC2CHW()
    type_cast_op = CT.TypeCast(mstype.int32)

    # apply map operations on images
    mnist_ds = mnist_ds.map(input_columns="label", operations=type_cast_op, num_parallel_workers=num_parallel_workers)
    mnist_ds = mnist_ds.map(input_columns="image", operations=resize_op, num_parallel_workers=num_parallel_workers)
    mnist_ds = mnist_ds.map(input_columns="image", operations=rescale_op, num_parallel_workers=num_parallel_workers)
    mnist_ds = mnist_ds.map(input_columns="image", operations=rescale_nml_op, num_parallel_workers=num_parallel_workers)
    mnist_ds = mnist_ds.map(input_columns="image", operations=hwc2chw_op, num_parallel_workers=num_parallel_workers)

    # apply DatasetOps
    buffer_size = 10000
    mnist_ds = mnist_ds.shuffle(buffer_size=buffer_size)  # 10000 as in LeNet train script
    mnist_ds = mnist_ds.batch(batch_size, drop_remainder=True)
    mnist_ds = mnist_ds.repeat(repeat_size)

    return mnist_ds


class LeNet5(nn.Cell):
    """
    Lenet network

    Args:
        num_class (int): Num classes. Default: 10.
        num_channel (int): Num channels. Default: 1.

    Returns:
        Tensor, output tensor
    Examples:
        >>> LeNet(num_class=10)

    """

    def __init__(self, num_class=10, num_channel=1):
        super(LeNet5, self).__init__()
        self.conv1 = nn.Conv2d(num_channel, 6, 5, pad_mode='valid')
        self.conv2 = nn.Conv2d(6, 16, 5, pad_mode='valid')
        self.fc1 = nn.Dense(16 * 5 * 5, 120, weight_init=Normal(0.02))
        self.fc2 = nn.Dense(120, 84, weight_init=Normal(0.02))
        self.fc3 = nn.Dense(84, num_class, weight_init=Normal(0.02))
        self.relu = nn.ReLU()
        self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
        self.flatten = nn.Flatten()

    def construct(self, x):
        x = self.max_pool2d(self.relu(self.conv1(x)))
        x = self.max_pool2d(self.relu(self.conv2(x)))
        x = self.flatten(x)
        x = self.relu(self.fc1(x))
        x = self.relu(self.fc2(x))
        x = self.fc3(x)
        return x


if __name__ == "__main__":
    context.set_context(mode=context.GRAPH_MODE, device_target="GPU")

    model_path = "./models/ckpt/mindspore_run/"
    os.system("rm -rf {0}*.ckpt {0}*.meta {0}*.pb".format(model_path))

    ds_train_path = "./datasets/MNIST_Data/train/"
    ds_train = create_dataset(ds_train_path, 32)

    network = LeNet5(10)
    net_loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction="mean")
    net_opt = nn.Momentum(network.trainable_params(), 0.01, 0.9)
    config_ck = CheckpointConfig(save_checkpoint_steps=1875, keep_checkpoint_max=5)
    ckpoint_cb = ModelCheckpoint(prefix="checkpoint_lenet", directory=model_path, config=config_ck)
    model = Model(network, net_loss, net_opt)

    print("============== Starting Training ==============")
    model.train(1, ds_train, callbacks=[LossMonitor(375), ckpoint_cb], dataset_sink_mode=True)
============== Starting Training ==============
epoch: 1 step: 375, loss is 2.2898183
epoch: 1 step: 750, loss is 2.2777305
epoch: 1 step: 1125, loss is 0.27802905
epoch: 1 step: 1500, loss is 0.032973606
epoch: 1 step: 1875, loss is 0.06105463

For details about how to obtain the MNIST dataset used in the example, see Downloading the Dataset. Use the PyNative mode for debugging, including the execution of single operator, common function, and network training model. For details, see Debugging in PyNative Mode.

Executing an Inference Model

Call the eval API of Model to implement inference. To facilitate model evaluation, you can set metrics when the Model API is initialized.

Metrics are used to evaluate models. Common metrics include Accuracy, Fbeta, Precision, Recall, and TopKCategoricalAccuracy. Generally, the comprehensive model quality cannot be evaluated by one model metric. Therefore, multiple metrics are often used together to evaluate the model.

Common built-in evaluation metrics are as follows:

  • Accuracy: evaluates a classification model. Generally, accuracy refers to the percentage of results correctly predicted by the model to all results. Formula: $\(Accuracy = (TP + TN)/(TP + TN + FP + FN)\)$

  • Precision: percentage of correctly predicted positive results to all predicted positive results. Formula: $\(Precision = TP/(TP + FP)\)$

  • Recall: percentage of correctly predicted positive results to all actual positive results. Formula: $\(Recall = TP/(TP + FN)\)$

  • Fbeta: harmonic mean of precision and recall.

Formula: $\(F_\beta = (1 + \beta^2) \cdot \frac{precisiont \cdot recall}{(\beta^2 \cdot precision) + recall}\)$

  • TopKCategoricalAccuracy: calculates the top K categorical accuracy.

A code example is as follows:

import mindspore.dataset as ds
import mindspore.dataset.transforms.c_transforms as CT
import mindspore.dataset.vision.c_transforms as CV
import mindspore.nn as nn
from mindspore import context, Model, load_checkpoint, load_param_into_net
from mindspore import dtype as mstype
from mindspore.common.initializer import Normal
from mindspore.dataset.vision import Inter
from mindspore.nn import Accuracy, Precision


class LeNet5(nn.Cell):
    """
    Lenet network

    Args:
        num_class (int): Num classes. Default: 10.
        num_channel (int): Num channels. Default: 1.

    Returns:
        Tensor, output tensor
    Examples:
        >>> LeNet(num_class=10)

    """

    def __init__(self, num_class=10, num_channel=1):
        super(LeNet5, self).__init__()
        self.conv1 = nn.Conv2d(num_channel, 6, 5, pad_mode='valid')
        self.conv2 = nn.Conv2d(6, 16, 5, pad_mode='valid')
        self.fc1 = nn.Dense(16 * 5 * 5, 120, weight_init=Normal(0.02))
        self.fc2 = nn.Dense(120, 84, weight_init=Normal(0.02))
        self.fc3 = nn.Dense(84, num_class, weight_init=Normal(0.02))
        self.relu = nn.ReLU()
        self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
        self.flatten = nn.Flatten()

    def construct(self, x):
        x = self.max_pool2d(self.relu(self.conv1(x)))
        x = self.max_pool2d(self.relu(self.conv2(x)))
        x = self.flatten(x)
        x = self.relu(self.fc1(x))
        x = self.relu(self.fc2(x))
        x = self.fc3(x)
        return x


def create_dataset(data_path, batch_size=32, repeat_size=1,
                   num_parallel_workers=1):
    """
    create dataset for train or test
    """
    # define dataset
    mnist_ds = ds.MnistDataset(data_path)

    resize_height, resize_width = 32, 32
    rescale = 1.0 / 255.0
    shift = 0.0
    rescale_nml = 1 / 0.3081
    shift_nml = -1 * 0.1307 / 0.3081

    # define map operations
    resize_op = CV.Resize((resize_height, resize_width), interpolation=Inter.LINEAR)  # Bilinear mode
    rescale_nml_op = CV.Rescale(rescale_nml, shift_nml)
    rescale_op = CV.Rescale(rescale, shift)
    hwc2chw_op = CV.HWC2CHW()
    type_cast_op = CT.TypeCast(mstype.int32)

    # apply map operations on images
    mnist_ds = mnist_ds.map(input_columns="label", operations=type_cast_op, num_parallel_workers=num_parallel_workers)
    mnist_ds = mnist_ds.map(input_columns="image", operations=resize_op, num_parallel_workers=num_parallel_workers)
    mnist_ds = mnist_ds.map(input_columns="image", operations=rescale_op, num_parallel_workers=num_parallel_workers)
    mnist_ds = mnist_ds.map(input_columns="image", operations=rescale_nml_op, num_parallel_workers=num_parallel_workers)
    mnist_ds = mnist_ds.map(input_columns="image", operations=hwc2chw_op, num_parallel_workers=num_parallel_workers)

    # apply DatasetOps
    buffer_size = 10000
    mnist_ds = mnist_ds.shuffle(buffer_size=buffer_size)  # 10000 as in LeNet train script
    mnist_ds = mnist_ds.batch(batch_size, drop_remainder=True)
    mnist_ds = mnist_ds.repeat(repeat_size)

    return mnist_ds


if __name__ == "__main__":
    context.set_context(mode=context.GRAPH_MODE, device_target="GPU")

    model_path = "./models/ckpt/mindspore_run/"
    ds_eval_path = "./datasets/MNIST_Data/test/"
    network = LeNet5(10)
    net_loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction="mean")
    repeat_size = 1
    net_opt = nn.Momentum(network.trainable_params(), 0.01, 0.9)
    model = Model(network, net_loss, net_opt, metrics={"Accuracy": Accuracy(), "Precision": Precision()})

    print("============== Starting Testing ==============")
    param_dict = load_checkpoint(model_path+"checkpoint_lenet-1_1875.ckpt")
    load_param_into_net(network, param_dict)
    ds_eval = create_dataset(ds_eval_path, 32, repeat_size)

    acc = model.eval(ds_eval, dataset_sink_mode=True)
    print("============== {} ==============".format(acc))
============== Starting Testing ==============
============== {'Accuracy': 0.960136217948718, 'Precision': array([0.95763547, 0.98059965, 0.99153439, 0.93333333, 0.97322348,
       0.99385749, 0.98502674, 0.93179724, 0.8974359 , 0.97148676])} ==============

In the preceding information:

  • load_checkpoint: loads the checkpoint model parameter file and returns a parameter dictionary.

  • checkpoint_lenet-1_1875.ckpt: name of the saved checkpoint model file.

  • load_param_into_net: loads parameters to the network.

For details about how to save the checkpoint_lenet-1_1875.ckpt file, see Training the Network.