microsoftml.featurize_image: 画像を特徴に変換する
使用法
microsoftml.featurize_image(cols: [dict, str], dnn_model: ['Resnet18',
'Resnet50', 'Resnet101', 'Alexnet'] = 'Resnet18', **kargs)
説明
事前トレーニング済みのディープ ニューラル ネットワークモデルを使用して、画像を特徴化します。
説明
featurize_image
では、指定された事前トレーニング済みのディープ ニューラル ネットワーク モデルを使用して、画像を特徴化します。 この変換に対する入力変数は、ピクセル値を抽出する必要があります。
引数
cols
抽出されたピクセル値を含む入力変数。
dict
の場合、キーは作成される新しい変数名を表します。
dnn_model
事前トレーニング済みのディープ ニューラル ネットワーク。 オプションは次のとおりです。
"Resnet18"
"Resnet50"
"Resnet101"
"Alexnet"
既定値は "Resnet18"
です。
ResNet の詳細については、「画像認識のための Deep Residual Learning」を参照してください。
kargs
コンピューティング エンジンに送信される追加の引数。
戻り値
変換を定義するオブジェクト。
関連項目
load_image
, resize_image
, extract_pixels
.
例
'''
Example with images.
'''
import numpy
import pandas
from microsoftml import rx_neural_network, rx_predict, rx_fast_linear
from microsoftml import load_image, resize_image, extract_pixels
from microsoftml.datasets.image import get_RevolutionAnalyticslogo
train = pandas.DataFrame(data=dict(Path=[get_RevolutionAnalyticslogo()], Label=[True]))
# Loads the images from variable Path, resizes the images to 1x1 pixels
# and trains a neural net.
model1 = rx_neural_network("Label ~ Features", data=train,
ml_transforms=[
load_image(cols=dict(Features="Path")),
resize_image(cols="Features", width=1, height=1, resizing="Aniso"),
extract_pixels(cols="Features")],
ml_transform_vars=["Path"],
num_hidden_nodes=1, num_iterations=1)
# Featurizes the images from variable Path using the default model, and trains a linear model on the result.
# If dnnModel == "AlexNet", the image has to be resized to 227x227.
model2 = rx_fast_linear("Label ~ Features ", data=train,
ml_transforms=[
load_image(cols=dict(Features="Path")),
resize_image(cols="Features", width=224, height=224),
extract_pixels(cols="Features")],
ml_transform_vars=["Path"], max_iterations=1)
# We predict even if it does not make too much sense on this single image.
print("\nrx_neural_network")
prediction1 = rx_predict(model1, data=train)
print(prediction1)
print("\nrx_fast_linear")
prediction2 = rx_predict(model2, data=train)
print(prediction2)
出力:
Automatically adding a MinMax normalization transform, use 'norm=Warn' or 'norm=No' to turn this behavior off.
Beginning processing data.
Rows Read: 1, Read Time: 0, Transform Time: 0
Beginning processing data.
Beginning processing data.
Rows Read: 1, Read Time: 0, Transform Time: 0
Beginning processing data.
Beginning processing data.
Rows Read: 1, Read Time: 0, Transform Time: 0
Beginning processing data.
Using: AVX Math
***** Net definition *****
input Data [3];
hidden H [1] sigmoid { // Depth 1
from Data all;
}
output Result [1] sigmoid { // Depth 0
from H all;
}
***** End net definition *****
Input count: 3
Output count: 1
Output Function: Sigmoid
Loss Function: LogLoss
PreTrainer: NoPreTrainer
___________________________________________________________________
Starting training...
Learning rate: 0.001000
Momentum: 0.000000
InitWtsDiameter: 0.100000
___________________________________________________________________
Initializing 1 Hidden Layers, 6 Weights...
Estimated Pre-training MeanError = 0.707823
Iter:1/1, MeanErr=0.707823(0.00%), 0.01M WeightUpdates/sec
Done!
Estimated Post-training MeanError = 0.707499
___________________________________________________________________
Not training a calibrator because it is not needed.
Elapsed time: 00:00:00.0751759
Elapsed time: 00:00:00.0080433
Automatically adding a MinMax normalization transform, use 'norm=Warn' or 'norm=No' to turn this behavior off.
Beginning processing data.
Rows Read: 1, Read Time: 0, Transform Time: 0
Beginning processing data.
Beginning processing data.
Rows Read: 1, Read Time: 0.001, Transform Time: 0
Beginning processing data.
Beginning processing data.
Rows Read: 1, Read Time: 0, Transform Time: 0
Beginning processing data.
Using 2 threads to train.
Automatically choosing a check frequency of 2.
Auto-tuning parameters: L2 = 5.
Auto-tuning parameters: L1Threshold (L1/L2) = 1.
Using model from last iteration.
Not training a calibrator because it is not needed.
Elapsed time: 00:00:01.0104773
Elapsed time: 00:00:00.0106935
rx_neural_network
Beginning processing data.
Rows Read: 1, Read Time: 0, Transform Time: 0
Beginning processing data.
Elapsed time: 00:00:00.0420328
Finished writing 1 rows.
Writing completed.
PredictedLabel Score Probability
0 False -0.028504 0.492875
rx_fast_linear
Beginning processing data.
Rows Read: 1, Read Time: 0, Transform Time: 0
Beginning processing data.
Elapsed time: 00:00:00.4449623
Finished writing 1 rows.
Writing completed.
PredictedLabel Score Probability
0 False 0.0 0.5