def load(self, model_path, batch_size=0):
"""
Load an openVINO model.
:param model_path: String. The file path to the OpenVINO IR xml file.
:param batch_size: Int. Set batch Size, default is 0 (use default batch size).
:return:
"""
self.node_num, self.core_num = get_node_and_core_number()
self.path = model_path
if batch_size != 0:
self.batch_size = batch_size
else:
import xml.etree.ElementTree as ET
tree = ET.parse(model_path)
root = tree.getroot()
shape_item = root.find('./layers/layer/output/port/dim[1]')
if shape_item is None:
raise ValueError(
"Invalid openVINO IR xml file, please check your model_path"
)
self.batch_size = int(shape_item.text)
self.model = InferenceModel(supported_concurrent_num=self.core_num)
self.model.load_openvino(
model_path=model_path,
weight_path=model_path[:model_path.rindex(".")] + ".bin",
batch_size=batch_size)
def test_load_caffe(self):
model = InferenceModel(10)
model.load_caffe(
os.path.join(resource_path, "models/caffe/test_persist.prototxt"),
os.path.join(resource_path,
"models/caffe/test_persist.caffemodel"))
input_data = np.random.random([4, 3, 8, 8])
output_data = model.predict(input_data)
def test_load_torch(self):
torch_model = torchvision.models.resnet18()
tmp_path = create_tmp_path() + ".pt"
torch.save(torch_model, tmp_path, pickle_module=zoo_pickle_module)
model = InferenceModel(10)
model.load_torch(tmp_path)
input_data = np.random.random([4, 3, 224, 224])
output_data = model.predict(input_data)
os.remove(tmp_path)
def load_model_bigdl(self):
fp = open("bigdl_model.model", 'wb')
fp.write(self.model)
fp.close()
from zoo.pipeline.inference import InferenceModel
model = InferenceModel()
model.load_bigdl("bigdl_model.model")
return model
def test_load_tf_openvino_ic(self):
local_path = self.create_temp_dir()
print(local_path)
url = data_url + "/models/resnet_v1_50_2016_08_28.tar.gz"
file_abs_path = maybe_download("resnet_v1_50_2016_08_28.tar.gz",
local_path, url)
tar = tarfile.open(file_abs_path, "r:gz")
print("Extracting %s to %s" % (file_abs_path, local_path))
tar.extractall(local_path)
tar.close()
model = InferenceModel(3)
model.load_tf_image_classification_as_openvino(
model_path=None,
image_classification_model_type="resnet_v1_50",
checkpoint_path=local_path + "/resnet_v1_50.ckpt",
input_shape=[4, 224, 224, 3],
if_reverse_input_channels=True,
mean_values=[123.68, 116.78, 103.94],
scale=1)
print(model)
input_data = np.random.random([4, 1, 224, 224, 3])
s3url = "https://s3-ap-southeast-1.amazonaws.com/"
var_url = s3url + "analytics-zoo-models/openvino/val_bmp_32.tar"
lib_url = s3url + "analytics-zoo-models/openvino/opencv_4.0.0_ubuntu_lib.tar"
var_file_abs_path = maybe_download("val_bmp_32.tar", local_path,
var_url)
lib_file_abs_path = maybe_download("opencv_4.0.0_ubuntu_lib.tar",
local_path, lib_url)
var_tar = tarfile.open(var_file_abs_path, "r")
print("Extracting %s to %s" % (var_file_abs_path, local_path))
var_tar.extractall(local_path)
var_tar.close()
lib_tar = tarfile.open(lib_file_abs_path, "r")
print("Extracting %s to %s" % (lib_file_abs_path, local_path))
lib_tar.extractall(local_path)
lib_tar.close()
validation_file_path = local_path + "/val_bmp_32/val.txt"
opencv_lib_path = local_path + "/lib"
model2 = InferenceModel(3)
model2.load_tf_as_calibrated_openvino(
model_path=None,
model_type="resnet_v1_50",
checkpoint_path=local_path + "/resnet_v1_50.ckpt",
input_shape=[4, 224, 224, 3],
if_reverse_input_channels=True,
mean_values=[123.68, 116.78, 103.94],
scale=1,
network_type='C',
validation_file_path=validation_file_path,
subset=32,
opencv_lib_path=opencv_lib_path)
print(model2)
model2.predict(input_data)
def test_load_openvino(self):
local_path = self.create_temp_dir()
url = data_url + "/IR_faster_rcnn_resnet101_coco_2018_01_28"
maybe_download("frozen_inference_graph.xml", local_path,
url + "/frozen_inference_graph.xml")
maybe_download("frozen_inference_graph.bin", local_path,
url + "/frozen_inference_graph.bin")
model = InferenceModel()
model.load_openvino(local_path + "/frozen_inference_graph.xml",
local_path + "/frozen_inference_graph.bin")
input_data = np.random.random([1, 1, 3, 600, 600])
output_data = model.predict(input_data)
def test_load_tf_openvino(self):
local_path = self.create_temp_dir()
url = data_url + "/models/object_detection/faster_rcnn_resnet101_coco_2018_01_28.tar.gz"
file_abs_path = maybe_download("faster_rcnn_resnet101_coco_2018_01_28.tar.gz",
local_path, url)
tar = tarfile.open(file_abs_path, "r:gz")
extracted_to = os.path.join(local_path, "faster_rcnn_resnet101_coco_2018_01_28")
if not os.path.exists(extracted_to):
print("Extracting %s to %s" % (file_abs_path, extracted_to))
tar.extractall(local_path)
tar.close()
model = InferenceModel(3)
model.load_tf(model_path=extracted_to + "/frozen_inference_graph.pb",
backend="openvino",
model_type="faster_rcnn_resnet101_coco",
ov_pipeline_config_path=extracted_to + "/pipeline.config",
ov_extensions_config_path=None)
input_data = np.random.random([4, 1, 3, 600, 600])
output_data = model.predict(input_data)
def __init__(self, *, model_path, batch_size=0):
self.node_num, self.core_num = get_node_and_core_number()
self.path = model_path
if batch_size != 0:
self.batch_size = batch_size
else:
import xml.etree.ElementTree as ET
tree = ET.parse(model_path)
root = tree.getroot()
shape_item = root.find('./layers/layer/output/port/dim[1]')
if shape_item is None:
raise ValueError(
"Invalid openVINO IR xml file, please check your model_path"
)
self.batch_size = int(shape_item.text)
self.model = InferenceModel(supported_concurrent_num=self.core_num)
self.model.load_openvino(
model_path=model_path,
weight_path=model_path[:model_path.rindex(".")] + ".bin",
batch_size=batch_size)
def predict(model_path, img_path):
model = InferenceModel()
model.load_openvino(model_path,
weight_path=model_path[:model_path.rindex(".")] +
".bin",
batch_size=BATCH_SIZE)
sc = init_nncontext("OpenVINO Python resnet_v1_50 Inference Example")
# pre-processing
infer_transformer = ChainedPreprocessing([
ImageBytesToMat(),
ImageResize(256, 256),
ImageCenterCrop(224, 224),
ImageMatToTensor(format="NHWC", to_RGB=True)
])
image_set = ImageSet.read(img_path, sc).\
transform(infer_transformer).get_image().collect()
image_set = np.expand_dims(image_set, axis=1)
for i in range(len(image_set) // BATCH_SIZE + 1):
index = i * BATCH_SIZE
# check whether out of index
if index >= len(image_set):
break
batch = image_set[index]
# put 4 images in one batch
for j in range(index + 1, min(index + BATCH_SIZE, len(image_set))):
batch = np.vstack((batch, image_set[j]))
batch = np.expand_dims(batch, axis=0)
# predict batch
predictions = model.predict(batch)
result = predictions[0]
# post-processing for Top-1
print("batch_" + str(i))
for r in result:
output = {}
max_index = np.argmax(r)
output["Top-1"] = str(max_index)
print("* Predict result " + str(output))
print("finished...")
sc.stop()
def test_load_tf_openvino(self):
local_path = self.create_temp_dir()
url = data_url + "/TF_faster_rcnn_resnet101_coco_2018_01_28"
maybe_download("frozen_inference_graph.pb", local_path,
url + "/frozen_inference_graph.pb")
maybe_download("pipeline.config", local_path, url + "/pipeline.config")
maybe_download("faster_rcnn_support.json", local_path,
url + "/faster_rcnn_support.json")
model = InferenceModel(3)
model.load_tf(local_path + "/frozen_inference_graph.pb",
backend="openvino",
ov_pipeline_config_path=local_path + "/pipeline.config",
ov_extensions_config_path=local_path +
"/faster_rcnn_support.json")
input_data = np.random.random([4, 1, 3, 600, 600])
output_data = model.predict(input_data)
model2 = InferenceModel(5)
model2.load_tf(local_path + "/frozen_inference_graph.pb",
backend="openvino",
model_type="faster_rcnn_resnet101_coco")
output_data2 = model2.predict(input_data)
def test_load_openvino(self):
local_path = self.create_temp_dir()
model = InferenceModel(1)
model_url = data_url + "/analytics-zoo-models/openvino/2018_R5/resnet_v1_50.xml"
weight_url = data_url + "/analytics-zoo-models/openvino/2018_R5/resnet_v1_50.bin"
model_path = maybe_download("resnet_v1_50.xml", local_path, model_url)
weight_path = maybe_download("resnet_v1_50.bin", local_path,
weight_url)
model.load_openvino(model_path, weight_path)
input_data = np.random.random([4, 1, 224, 224, 3])
model.predict(input_data)
if __name__ == "__main__":
parser = OptionParser()
parser.add_option("--image",
type=str,
dest="img_path",
help="The path where the images are stored, "
"can be either a folder or an image path")
parser.add_option("--model",
type=str,
dest="model_path",
help="Path to the TensorFlow model file")
(options, args) = parser.parse_args(sys.argv)
sc = init_nncontext("OpenVINO Object Detection Inference Example")
images = ImageSet.read(options.img_path,
sc,
resize_height=600,
resize_width=600).get_image().collect()
input_data = np.concatenate(
[image.reshape((1, 1) + image.shape) for image in images], axis=0)
model_path = options.model_path
model = InferenceModel()
model.load_openvino(model_path,
weight_path=model_path[:model_path.rindex(".")] +
".bin")
predictions = model.predict(input_data)
# Print the detection result of the first image.
print(predictions[0])
def classify_process(model_path):
print("* Loading model...")
model = InferenceModel()
model.load_openvino(model_path=model_path,
weight_path=model_path[:model_path.rindex(".")] +
".bin")
print("* Model loaded")
# continually poll for new images to classify
while True:
image_ids = []
batch = None
# loop over the queue
start_time = time.time()
count = 0
while count < 4:
# Get message or None or timeout
record = pub.get_message()
if record and record['type'] == 'message':
data = json.loads(record['data'].decode("utf-8"))
image = helpers.base64_decode_image(data["image"])
image = helpers.byte_to_mat(image, dtype=settings.IMAGE_DTYPE)
image = helpers.image_preprocess(image, settings.IMAGE_WIDTH,
settings.IMAGE_HEIGHT)
# check to see if the batch list is None
if batch is None:
batch = image
# otherwise, stack the data
else:
batch = np.vstack([batch, image])
# update the list of image IDs
image_ids.append(data["id"])
count += 1
print("* Pop from redis %d ms" %
int(round((time.time() - start_time) * 1000)))
# check to see if we need to process the batch
if len(image_ids) > 0:
# classify the batch
batch = np.expand_dims(batch, axis=0)
print("* Batch size: {}".format(batch.shape))
# Output is [1, 4, 1000]
results = model.predict(batch)[0]
print("* Predict a batch %d ms" %
int(round((time.time() - start_time) * 1000)))
# loop over the image IDs and their corresponding set of
# results from our model
for (imageID, resultSet) in zip(image_ids, results):
# initialize the list of output predictions
output = {}
# loop over the results and add them to the list of
# output predictions
# Top 1
max_index = np.argmax(resultSet)
output["Top-1"] = str(max_index)
output["id"] = imageID
print("* Predict result " + str(output))
# store the output predictions in the database, using
# the image ID as the key so we can fetch the results
DB.lpush(settings.PREDICT_QUEUE, json.dumps(output))
# remove the set of images from our queue
print("* Total time used is %d ms" %
int(round((time.time() - start_time) * 1000)))
# sleep for a small amount
time.sleep(settings.SERVER_SLEEP)
class OpenvinoEstimatorWrapper(SparkEstimator):
def __init__(self, *, model_path, batch_size=0):
self.node_num, self.core_num = get_node_and_core_number()
self.path = model_path
if batch_size != 0:
self.batch_size = batch_size
else:
import xml.etree.ElementTree as ET
tree = ET.parse(model_path)
root = tree.getroot()
shape_item = root.find('./layers/layer/output/port/dim[1]')
if shape_item is None:
raise ValueError(
"Invalid openVINO IR xml file, please check your model_path"
)
self.batch_size = int(shape_item.text)
self.model = InferenceModel(supported_concurrent_num=self.core_num)
self.model.load_openvino(
model_path=model_path,
weight_path=model_path[:model_path.rindex(".")] + ".bin",
batch_size=batch_size)
def fit(self, data, epochs, **kwargs):
raise NotImplementedError
def predict(self, data, **kwargs):
def predict_transform(dict_data, batch_size):
assert isinstance(dict_data, dict), "each shard should be an dict"
assert "x" in dict_data, "key x should in each shard"
feature_data = dict_data["x"]
if isinstance(feature_data, np.ndarray):
assert feature_data.shape[1] <= batch_size, \
"The batch size of input data (the second dim) should be less than the model " \
"batch size, otherwise some inputs will be ignored."
elif isinstance(feature_data, list):
for elem in feature_data:
assert isinstance(elem, np.ndarray), "Each element in the x list should be " \
"a ndarray, but get " + \
elem.__class__.__name__
assert elem.shape[1] <= batch_size, "The batch size of each input data (the " \
"second dim) should be less than the " \
"model batch size, otherwise some inputs " \
"will be ignored."
else:
raise ValueError(
"x in each shard should be a ndarray or a list of ndarray."
)
return dict_data["x"]
sc = init_nncontext()
if isinstance(data, SparkXShards):
assert sc is not None, "You should pass sc(spark context) if data is a XShards."
from zoo.orca.learn.utils import convert_predict_to_xshard
data = data.transform_shard(predict_transform, self.batch_size)
result_rdd = self.model.distributed_predict(data.rdd, sc)
return convert_predict_to_xshard(result_rdd)
elif isinstance(data, (np.ndarray, list)):
total_core_num = self.core_num * self.node_num
if isinstance(data, np.ndarray):
assert data.shape[1] <= self.batch_size, "The batch size of input data (the " \
"second dim) should be less than the " \
"model batch size, otherwise some " \
"inputs will be ignored."
split_num = min(total_core_num, data.shape[0])
arrays = np.array_split(data, split_num)
data_rdd = sc.parallelize(arrays, numSlices=split_num)
elif isinstance(data, list):
flattened = nest.flatten(data)
data_length = len(flattened[0])
data_to_be_rdd = []
split_num = min(total_core_num, flattened[0].shape[0])
for i in range(split_num):
data_to_be_rdd.append([])
for x in flattened:
assert isinstance(x, np.ndarray), "the data in the data list should be " \
"ndarrays, but get " + \
x.__class__.__name__
assert len(x) == data_length, \
"the ndarrays in data must all have the same size in first dimension" \
", got first ndarray of size {} and another {}".format(data_length, len(x))
assert x.shape[1] <= self.batch_size, "The batch size of each input data (" \
"the second dim) should be less than " \
"the model batch size, otherwise some " \
"inputs will be ignored."
x_parts = np.array_split(x, split_num)
for idx, x_part in enumerate(x_parts):
data_to_be_rdd[idx].append(x_part)
data_to_be_rdd = [
nest.pack_sequence_as(data, shard)
for shard in data_to_be_rdd
]
data_rdd = sc.parallelize(data_to_be_rdd, numSlices=split_num)
result_rdd = self.model.distributed_predict(data_rdd, sc)
result_arr_list = result_rdd.collect()
result_arr = np.concatenate(result_arr_list, axis=0)
return result_arr
else:
raise ValueError(
"Only XShards, a numpy array and a list of numpy arrays are supported "
"as input data, but get " + data.__class__.__name__)
def evaluate(self, data, **kwargs):
raise NotImplementedError
def get_model(self):
raise NotImplementedError
def save(self, model_path):
raise NotImplementedError
def load(self, model_path, batch_size=0):
"""
Load an openVINO model.
:param model_path: String. The file path to the OpenVINO IR xml file.
:param batch_size: Int. Set batch Size, default is 0 (use default batch size).
:return:
"""
self.node_num, self.core_num = get_node_and_core_number()
self.path = model_path
if batch_size != 0:
self.batch_size = batch_size
else:
import xml.etree.ElementTree as ET
tree = ET.parse(model_path)
root = tree.getroot()
shape_item = root.find('./layers/layer/output/port/dim[1]')
if shape_item is None:
raise ValueError(
"Invalid openVINO IR xml file, please check your model_path"
)
self.batch_size = int(shape_item.text)
self.model = InferenceModel(supported_concurrent_num=self.core_num)
self.model.load_openvino(
model_path=model_path,
weight_path=model_path[:model_path.rindex(".")] + ".bin",
batch_size=batch_size)
def set_tensorboard(self, log_dir, app_name):
raise NotImplementedError
def clear_gradient_clipping(self):
raise NotImplementedError
def set_constant_gradient_clipping(self, min, max):
raise NotImplementedError
def set_l2_norm_gradient_clipping(self, clip_norm):
raise NotImplementedError
def get_train_summary(self, tag=None):
raise NotImplementedError
def get_validation_summary(self, tag=None):
raise NotImplementedError
def load_orca_checkpoint(self, path, version):
raise NotImplementedError
def load_latest_orca_checkpoint(self, path):
raise NotImplementedError
import numpy as np
from optparse import OptionParser
from zoo.common.nncontext import init_nncontext
from zoo.feature.image import ImageSet
from zoo.pipeline.inference import InferenceModel
if __name__ == "__main__":
parser = OptionParser()
parser.add_option("--image", type=str, dest="img_path",
help="The path where the images are stored, "
"can be either a folder or an image path")
parser.add_option("--model", type=str, dest="model_path",
help="Path to the TensorFlow model file")
parser.add_option("--model_type", type=str, dest="model_type",
help="The type of the TensorFlow model",
default="faster_rcnn_resnet101_coco")
(options, args) = parser.parse_args(sys.argv)
sc = init_nncontext("OpenVINO Object Detection Inference Example")
images = ImageSet.read(options.img_path, sc,
resize_height=600, resize_width=600).get_image().collect()
input_data = np.concatenate([image.reshape((1, 1) + image.shape) for image in images], axis=0)
model = InferenceModel()
model.load_tf(options.model_path, backend="openvino", model_type=options.model_type)
predictions = model.predict(input_data)
# Print the detection result of the first image.
print(predictions[0])
help="The path where the images are stored, "
"can be either a folder or an image path")
parser.add_option("--model",
type=str,
dest="model_path",
help="Path to the TensorFlow model file")
parser.add_option("--model_type",
type=str,
dest="model_type",
help="The type of the TensorFlow model",
default="faster_rcnn_resnet101_coco")
(options, args) = parser.parse_args(sys.argv)
sc = init_nncontext("OpenVINO Object Detection Inference Example")
images = ImageSet.read(options.img_path,
sc,
resize_height=600,
resize_width=600).get_image().collect()
input_data = np.concatenate(
[image.reshape((1, 1) + image.shape) for image in images], axis=0)
model = InferenceModel()
model.load_tf(join(options.model_path, "frozen_inference_graph.pb"),
backend="openvino",
model_type=options.model_type,
ov_pipeline_config_path=join(options.model_path,
"pipeline.config"))
predictions = model.predict(input_data)
# Print the detection result of the first image.
print(predictions[0])
def classify_process(model_path):
# load the pre-trained OpenVINO model
print("* Loading model...")
model = InferenceModel()
model.load_openvino(model_path,
weight_path=model_path[:model_path.rindex(".")] +
".bin")
print("* Model loaded")
# continually pool for new images to classify
while True:
# attempt to grab a batch of images from the database, then
# initialize the image IDs and batch of images themselves
queue = DB.lrange(settings.IMAGE_QUEUE, 0, settings.BATCH_SIZE - 1)
image_ids = []
batch = None
# loop over the queue
start_time = time.time()
for q in queue:
# deserialize the object and obtain the input image
q = json.loads(q.decode("utf-8"))
image = helpers.base64_decode_image(q["image"],
settings.IMAGE_DTYPE)
# check to see if the batch list is None
if batch is None:
batch = image
# otherwise, stack the data
else:
batch = np.vstack([batch, image])
# update the list of image IDs
image_ids.append(q["id"])
print("* Pop from redis %d ms" %
int(round((time.time() - start_time) * 1000)))
# check to see if we need to process the batch
if len(image_ids) > 0:
# classify the batch
batch = np.expand_dims(batch, axis=0)
print("* Batch size: {}".format(batch.shape))
# Output is [1, 4, 1000]
results = model.predict(batch)[0]
print("* Predict a batch %d ms" %
int(round((time.time() - start_time) * 1000)))
# loop over the image IDs and their corresponding set of
# results from our model
for (imageID, resultSet) in zip(image_ids, results):
# initialize the list of output predictions
output = {}
# loop over the results and add them to the list of
# output predictions
# Top 1
max_index = np.argmax(resultSet)
output["Top-1"] = str(max_index)
output["id"] = imageID
print("* Predict result " + str(output))
# store the output predictions in the database, using
# the image ID as the key so we can fetch the results
DB.lpush(settings.PREDICT_QUEUE, json.dumps(output))
# remove the set of images from our queue
print("* Total time used is %d ms" %
int(round((time.time() - start_time) * 1000)))
DB.ltrim(settings.IMAGE_QUEUE, len(image_ids), -1)
# sleep for a small amount
time.sleep(settings.SERVER_SLEEP)
def test_load_bigdl(self):
model = InferenceModel(3)
model.load_bigdl(
os.path.join(resource_path, "models/bigdl/bigdl_lenet.model"))
input_data = np.random.random([4, 28, 28, 1])
output_data = model.predict(input_data)
class OpenvinoEstimator(SparkEstimator):
def __init__(self, *, model_path, batch_size=0):
self.node_num, self.core_num = get_node_and_core_number()
self.path = model_path
if batch_size != 0:
self.batch_size = batch_size
else:
import xml.etree.ElementTree as ET
tree = ET.parse(model_path)
root = tree.getroot()
shape_item = root.find('./layers/layer/output/port/dim[1]')
if shape_item is None:
raise ValueError(
"Invalid openVINO IR xml file, please check your model_path"
)
self.batch_size = int(shape_item.text)
self.model = InferenceModel(supported_concurrent_num=self.core_num)
self.model.load_openvino_ng(
model_path=model_path,
weight_path=model_path[:model_path.rindex(".")] + ".bin",
batch_size=self.batch_size)
def fit(self,
data,
epochs,
batch_size=32,
feature_cols=None,
label_cols=None,
validation_data=None,
checkpoint_trigger=None):
"""
Fit is not supported in OpenVINOEstimator
"""
raise NotImplementedError
def predict(self, data, feature_cols=None):
"""
Predict input data
:param data: data to be predicted. XShards, Spark DataFrame, numpy array and list of numpy
arrays are supported. If data is XShards, each partition is a dictionary of {'x':
feature}, where feature(label) is a numpy array or a list of numpy arrays.
:param feature_cols: Feature column name(s) of data. Only used when data is a Spark
DataFrame. Default: None.
:return: predicted result.
If the input data is XShards, the predict result is a XShards, each partition
of the XShards is a dictionary of {'prediction': result}, where the result is a
numpy array or a list of numpy arrays.
If the input data is numpy arrays or list of numpy arrays, the predict result is
a numpy array or a list of numpy arrays.
"""
from pyspark.sql import DataFrame
def predict_transform(dict_data, batch_size):
assert isinstance(dict_data, dict), "each shard should be an dict"
assert "x" in dict_data, "key x should in each shard"
feature_data = dict_data["x"]
if isinstance(feature_data, np.ndarray):
assert feature_data.shape[0] <= batch_size, \
"The batch size of input data (the second dim) should be less than the model " \
"batch size, otherwise some inputs will be ignored."
elif isinstance(feature_data, list):
for elem in feature_data:
assert isinstance(elem, np.ndarray), "Each element in the x list should be " \
"a ndarray, but get " + \
elem.__class__.__name__
assert elem.shape[0] <= batch_size, "The batch size of each input data (the " \
"second dim) should be less than the " \
"model batch size, otherwise some inputs " \
"will be ignored."
else:
raise ValueError(
"x in each shard should be a ndarray or a list of ndarray."
)
return feature_data
sc = init_nncontext()
if isinstance(data, DataFrame):
from zoo.orca.learn.utils import dataframe_to_xshards, convert_predict_rdd_to_dataframe
xshards, _ = dataframe_to_xshards(data,
validation_data=None,
feature_cols=feature_cols,
label_cols=None,
mode="predict")
transformed_data = xshards.transform_shard(predict_transform,
self.batch_size)
result_rdd = self.model.distributed_predict(
transformed_data.rdd, sc)
def delete_useless_result(data):
shard, y = data
data_length = len(shard["x"])
return y[:data_length]
result_rdd = xshards.rdd.zip(result_rdd).map(delete_useless_result)
return convert_predict_rdd_to_dataframe(
data, result_rdd.flatMap(lambda data: data))
elif isinstance(data, SparkXShards):
transformed_data = data.transform_shard(predict_transform,
self.batch_size)
result_rdd = self.model.distributed_predict(
transformed_data.rdd, sc)
def update_shard(data):
shard, y = data
data_length = len(shard["x"])
shard["prediction"] = y[:data_length]
return shard
return SparkXShards(data.rdd.zip(result_rdd).map(update_shard))
elif isinstance(data, (np.ndarray, list)):
if isinstance(data, np.ndarray):
split_num = math.ceil(len(data) / self.batch_size)
arrays = np.array_split(data, split_num)
data_length_list = list(map(lambda arr: len(arr), arrays))
data_rdd = sc.parallelize(arrays, numSlices=split_num)
elif isinstance(data, list):
flattened = nest.flatten(data)
data_length = len(flattened[0])
data_to_be_rdd = []
split_num = math.ceil(flattened[0].shape[0] / self.batch_size)
for i in range(split_num):
data_to_be_rdd.append([])
for x in flattened:
assert isinstance(x, np.ndarray), "the data in the data list should be " \
"ndarrays, but get " + \
x.__class__.__name__
assert len(x) == data_length, \
"the ndarrays in data must all have the same size in first dimension" \
", got first ndarray of size {} and another {}".format(data_length, len(x))
x_parts = np.array_split(x, split_num)
for idx, x_part in enumerate(x_parts):
data_to_be_rdd[idx].append(x_part)
data_length_list = list(
map(lambda arr: len(arr), x_part))
data_to_be_rdd = [
nest.pack_sequence_as(data, shard)
for shard in data_to_be_rdd
]
data_rdd = sc.parallelize(data_to_be_rdd, numSlices=split_num)
result_rdd = self.model.distributed_predict(data_rdd, sc)
result_arr_list = result_rdd.collect()
for i in range(0, len(result_arr_list)):
result_arr_list[i] = result_arr_list[i][:data_length_list[i]]
result_arr = np.concatenate(result_arr_list, axis=0)
return result_arr
else:
raise ValueError(
"Only XShards, Spark DataFrame, a numpy array and a list of numpy arr"
"ays are supported as input data, but get " +
data.__class__.__name__)
def evaluate(self,
data,
batch_size=32,
feature_cols=None,
label_cols=None):
"""
Evaluate is not supported in OpenVINOEstimator
"""
raise NotImplementedError
def get_model(self):
"""
Get_model is not supported in OpenVINOEstimator
"""
raise NotImplementedError
def save(self, model_path):
"""
Save is not supported in OpenVINOEstimator
"""
raise NotImplementedError
def load(self, model_path, batch_size=0):
"""
Load an openVINO model.
:param model_path: String. The file path to the OpenVINO IR xml file.
:param batch_size: Int. Set batch Size, default is 0 (use default batch size).
:return:
"""
self.node_num, self.core_num = get_node_and_core_number()
self.path = model_path
if batch_size != 0:
self.batch_size = batch_size
else:
import xml.etree.ElementTree as ET
tree = ET.parse(model_path)
root = tree.getroot()
shape_item = root.find('./layers/layer/output/port/dim[1]')
if shape_item is None:
raise ValueError(
"Invalid openVINO IR xml file, please check your model_path"
)
self.batch_size = int(shape_item.text)
self.model = InferenceModel(supported_concurrent_num=self.core_num)
self.model.load_openvino(
model_path=model_path,
weight_path=model_path[:model_path.rindex(".")] + ".bin",
batch_size=batch_size)
def set_tensorboard(self, log_dir, app_name):
"""
Set_tensorboard is not supported in OpenVINOEstimator
"""
raise NotImplementedError
def clear_gradient_clipping(self):
"""
Clear_gradient_clipping is not supported in OpenVINOEstimator
"""
raise NotImplementedError
def set_constant_gradient_clipping(self, min, max):
"""
Set_constant_gradient_clipping is not supported in OpenVINOEstimator
"""
raise NotImplementedError
def set_l2_norm_gradient_clipping(self, clip_norm):
"""
Set_l2_norm_gradient_clipping is not supported in OpenVINOEstimator
"""
raise NotImplementedError
def get_train_summary(self, tag=None):
"""
Get_train_summary is not supported in OpenVINOEstimator
"""
raise NotImplementedError
def get_validation_summary(self, tag=None):
"""
Get_validation_summary is not supported in OpenVINOEstimator
"""
raise NotImplementedError
def load_orca_checkpoint(self, path, version):
"""
Load_orca_checkpoint is not supported in OpenVINOEstimator
"""
raise NotImplementedError
# store the output predictions in the database, using
# the image ID as the key so we can fetch the results
DB.lpush(settings.PREDICT_QUEUE, json.dumps(output))
# remove the set of images from our queue
DB.ltrim(settings.IMAGE_QUEUE, len(image_ids), -1)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--model_path', help="Zoo model path")
args = parser.parse_args()
model_path = args.model_path
print("* Loading model...")
model = InferenceModel()
model.load_openvino(model_path=model_path,
weight_path=model_path[:model_path.rindex(".")] +
".bin")
print("* Model loaded")
spark = SparkSession \
.builder \
.appName("Streaming Image Consumer") \
.config("spark.redis.host", settings.REDIS_HOST) \
.config("spark.redis.port", settings.REDIS_PORT) \
.getOrCreate()
# Streaming schema
imageSchema = StructType().add("id", "string").add("path", "string").add(
"image", "string")
def load_model(model_path, weight_path, batch_size):
model = InferenceModel(supported_concurrent_num=1)
model.load_openvino(model_path=model_path,
weight_path=weight_path,
batch_size=batch_size)
return model
