Usage
microsoftml.rx_neural_network(formula: str,
data: [revoscalepy.datasource.RxDataSource.RxDataSource,
pandas.core.frame.DataFrame], method: ['binary', 'multiClass',
'regression'] = 'binary', num_hidden_nodes: int = 100,
num_iterations: int = 100,
optimizer: [<function adadelta_optimizer at 0x0000007156EAC048>,
<function sgd_optimizer at 0x0000007156E9FB70>] = {'Name': 'SgdOptimizer',
'Settings': {}}, net_definition: str = None,
init_wts_diameter: float = 0.1, max_norm: float = 0,
acceleration: [<function avx_math at 0x0000007156E9FEA0>,
<function clr_math at 0x0000007156EAC158>,
<function gpu_math at 0x0000007156EAC1E0>,
<function mkl_math at 0x0000007156EAC268>,
<function sse_math at 0x0000007156EAC2F0>] = {'Name': 'AvxMath',
'Settings': {}}, mini_batch_size: int = 1, normalize: ['No',
'Warn', 'Auto', 'Yes'] = 'Auto', ml_transforms: list = None,
ml_transform_vars: list = None, row_selection: str = None,
transforms: dict = None, transform_objects: dict = None,
transform_function: str = None,
transform_variables: list = None,
transform_packages: list = None,
transform_environment: dict = None, blocks_per_read: int = None,
report_progress: int = None, verbose: int = 1,
ensemble: microsoftml.modules.ensemble.EnsembleControl = None,
compute_context: revoscalepy.computecontext.RxComputeContext.RxComputeContext = None)
Description
회귀 모델링 및 이진 및 다중 클래스 분류를 위한 신경망.
세부 정보
신경망은 인간의 뇌에서 영감을 얻은 예측 모델의 클래스입니다. 신경망은 가중치를 지정한 그래프로 나타낼 수 있습니다. 그래프의 각 노드를 뉴런이라고합니다. 그래프의 뉴런은 계층으로 정렬됩니다. 여기서 한 레이어의 뉴런은 가중치가 적용된 가장자리(가중치는 0 또는 양수일 수 있습니다)를 통해 다음 레이어의 뉴런에 연결됩니다. 첫 번째 계층을 입력 계층이라고 하며 입력 계층의 각 뉴런은 기능 중 하나에 해당합니다. 함수의 마지막 계층을 출력 계층이라고 합니다. 따라서 이진 신경망의 경우 각 클래스에 대해 하나씩 두 개의 출력 뉴런이 포함되며, 값은 각 클래스에 속할 확률입니다. 나머지 레이어를 숨겨진 레이어라고 합니다. 숨겨진 계층 및 출력 계층의 뉴런 값은 이전 계층에서 뉴런 값의 가중 합계를 계산하고 해당 가중 합계에 활성화 함수를 적용하여 설정됩니다. 신경망 모델은 그래프의 구조(즉, 숨겨진 레이어의 수와 숨겨진 각 계층의 뉴런 수), 활성화 함수 선택 및 그래프 가장자리의 가중치로 정의됩니다. 신경망 알고리즘은 학습 데이터를 기반으로 에지에서 최적 가중치를 학습하려고 합니다.
신경망은 이미지 인식과 같은 복잡한 문제를 심층 학습 및 모델링하는 데 널리 사용되지만 회귀 문제에 쉽게 적응할 수 있습니다. 통계 모델의 모든 클래스는 적응 가중치를 사용하고 입력의 비선형 함수를 근사화할 수 있는 경우 신경망으로 간주될 수 있습니다. 신경망 회귀는 보다 전통적인 회귀 모델이 솔루션에 맞지 않는 문제에 특히 적합합니다.
Arguments
식
revoscalepy.rx_formula 설명된 수식입니다.
상호 작용 용어이며 F() 현재 microsoftml에서 지원되지 않습니다.
데이터
데이터 원본 개체 또는 .xdf 파일 또는 데이터 프레임 개체를 지정하는 문자열입니다.
method
빠른 트리 형식을 나타내는 문자열입니다.
"binary"기본 이진 분류 신경망에 대한 입니다."multiClass"다중 클래스 분류 신경망의 경우"regression"회귀 신경망의 경우
num_hidden_nodes
신경망에 숨겨진 노드의 기본 수입니다. 기본값은 100입니다.
num_iterations
전체 학습 집합의 반복 횟수입니다. 기본값은 100입니다.
optimizer
최적화 알고리즘을 adaptive 지정하는 sgd 목록입니다. 이 목록은 사용하거나 adadelta_optimizer.를 사용하여 sgd_optimizer 만들 수 있습니다.
기본값은 sgd입니다.
net_definition
신경망 구조의 Net# 정의입니다. Net# 언어에 대한 자세한 내용은 참조 가이드를 참조하세요.
init_wts_diameter
초기 학습 가중치에 대해 값을 그릴 범위를 지정하는 초기 가중치 지름을 설정합니다. 가중치는 이 범위 내에서 임의로 초기화됩니다. 기본값은 0.1입니다.
max_norm
각 숨겨진 단위에서 들어오는 가중치 벡터의 표준을 제한하기 위해 상한을 지정합니다. 이는 학습이 무제한 가중치를 생성하는 경우뿐만 아니라 최대 아웃 신경망에서 매우 중요할 수 있습니다.
가속
사용할 하드웨어 가속 유형을 지정합니다. 가능한 값은 "sse_math" 및 "gpu_math"입니다. GPU 가속의 경우 1보다 큰 miniBatchSize를 사용하는 것이 좋습니다. GPU 가속을 사용하려면 추가 수동 설정 단계가 필요합니다.
NVidia CUDA Toolkit 6.5(CUDA 도구 키트)를 다운로드하여 설치합니다.
NVidia cuDNN v2 라이브러리(cudnn 라이브러리)를 다운로드하여 설치합니다.
를 호출
import microsoftml, osos.path.join(microsoftml.__path__[0], "mxLibs")하여 microsoftml 패키지의 libs 디렉터리를 찾습니다.CUDA 도구 키트 6.5의 cublas64_65.dll, cudart64_65.dll및cusparse64_65.dll microsoftml 패키지의 libs 디렉터리에 복사합니다.
cuDNN v2 라이브러리의 cudnn64_65.dll microsoftml 패키지의 libs 디렉터리에 복사합니다.
mini_batch_size
미니 일괄 처리 크기를 설정합니다. 권장 값은 1에서 256 사이입니다. 이 매개 변수는 가속이 GPU인 경우에만 사용됩니다. 이 매개 변수를 더 높은 값으로 설정하면 학습 속도가 향상되지만 정확도에 부정적인 영향을 줄 수 있습니다. 기본값은 1입니다.
정규화
사용되는 자동 정규화 유형을 지정합니다.
"Warn": 정규화가 필요한 경우 자동으로 수행됩니다. 이것이 기본 선택입니다."No": 정규화가 수행되지 않습니다."Yes": 정규화가 수행됩니다."Auto": 정규화가 필요한 경우 경고 메시지가 표시되지만 정규화는 수행되지 않습니다.
정규화는 서로 다른 데이터 범위를 표준 크기 조정으로 다시 크기 조정합니다. 기능 크기 조정은 데이터 요소 간의 거리가 비례하도록 보장하고 그라데이션 하강과 같은 다양한 최적화 방법을 사용하여 훨씬 더 빠르게 수렴할 수 있도록 합니다. 정규화가 수행되면 MaxMin 정규화기가 사용됩니다. 간격 [a, b]의 위치 -1 <= a <= 0 및 0 <= b <= 1b - a = 1. 이 정규화기는 0에서 0으로 매핑하여 스파스를 유지합니다.
ml_transforms
학습 전에 데이터에 대해 수행할 MicrosoftML 변환 목록을 지정하거나 변환을 수행할 항목이 없으면 None 을 지정합니다. 지원되는 변환은 , categorical및 categorical_hash,를 참조featurize_text하세요.
이러한 변환은 지정된 Python 변환 후에 수행됩니다.
기본 값은 None입니다.
ml_transform_vars
사용할 변수 이름의 문자 벡터를 지정하거나 사용할 ml_transforms 변수가 없는 경우 None 을 지정합니다.
기본 값은 None입니다.
row_selection
지원되지 않습니다. 데이터 집합의 논리 변수 이름(따옴표)을 사용하거나 데이터 집합의 변수를 사용하는 논리 식과 함께 모델에서 사용할 데이터 집합의 행(관찰)을 지정합니다. 다음은 그 예입니다.
row_selection = "old"는 변수oldTrue값이 있는 관찰만 사용합니다.row_selection = (age > 20) & (age < 65) & (log(income) > 10)변수 값이 20에서 65logincome사이이고 변수 값age이 10보다 큰 관찰만 사용합니다.
행 선택은 데이터 변환을 처리한 후 수행됩니다(인수 transforms 또는 transform_function참조). 모든 식과 마찬가지로 함수 row_selection 를 사용하여 expression 함수 호출 외부에서 정의할 수 있습니다.
변환
지원되지 않습니다. 변수 변환의 첫 번째 라운드를 나타내는 폼의 식입니다. 모든 식과 마찬가지로 함수를 transforms 사용하여 expression 함수 호출 외부에서 정의row_selection할 수 있습니다.
transform_objects
지원되지 않습니다. , transform_function및 row_selection.에서 참조할 수 있는 개체가 포함된 명명된 transforms목록입니다.
transform_function
변수 변환 함수입니다.
transform_variables
변환 함수에 필요한 입력 데이터 집합 변수의 문자 벡터입니다.
transform_packages
지원되지 않습니다. 가변 변환 함수에 RxOptions.get_option("transform_packages")사용할 수 있도록 미리 로드할 추가 Python 패키지(지정된 패키지 외부)를 지정하는 문자 벡터입니다.
예를 들어 해당 함수와 transform_function 인수를 통해 transformsrevoscalepy 함수에 명시적으로 정의되거나 해당 함수 또는 row_selection 인수를 통해 formula 암시적으로 정의된 함수입니다. 인수는 transform_packages외부RxOptions.get_option("transform_packages") 패키지가 미리 로드되지 않았음을 나타내는 None일 수도 있습니다.
transform_environment
지원되지 않습니다. 내부적으로 개발되어 변수 데이터 변환에 사용되는 모든 환경의 부모 역할을 하는 사용자 정의 환경입니다.
이 경우 transform_environment = None부모 revoscalepy.baseenvis가 대신 사용되는 새 "해시" 환경입니다.
blocks_per_read
데이터 원본에서 읽은 데이터의 각 청크에 대해 읽을 블록 수를 지정합니다.
report_progress
행 처리 진행률에 대한 보고 수준을 지정하는 정수 값입니다.
0: 진행률이 보고되지 않습니다.1: 처리된 행 수가 인쇄되고 업데이트됩니다.2: 행이 처리되고 타이밍이 보고됩니다.3: 행이 처리되고 모든 타이밍이 보고됩니다.
verbose
원하는 출력의 양을 지정하는 정수 값입니다.
이 경우 0계산 중에 자세한 정보 출력이 인쇄되지 않습니다. 증가하는 양의 정보를 제공하기 위한 4 정수 값 1 입니다.
compute_context
계산이 실행되고 유효한 revoscalepy로 지정된 컨텍스트를 설정합니다. RxComputeContext. 현재 로컬 및 revoscalepy입니다. RxInSqlServer 컴퓨팅 컨텍스트가 지원됩니다.
앙상블
ensembling에 대한 컨트롤 매개 변수입니다.
반품
NeuralNetwork 학습된 모델이 있는 개체입니다.
비고
이 알고리즘은 단일 스레드이며 전체 데이터 세트를 메모리에 로드하려고 시도하지 않습니다.
참고하십시오
adadelta_optimizer,sgd_optimizer, avx_math, clr_math, gpu_mathmkl_math, sse_math. rx_predict
References
이진 분류 예제
'''
Binary Classification.
'''
import numpy
import pandas
from microsoftml import rx_neural_network, rx_predict
from revoscalepy.etl.RxDataStep import rx_data_step
from microsoftml.datasets.datasets import get_dataset
infert = get_dataset("infert")
import sklearn
if sklearn.__version__ < "0.18":
from sklearn.cross_validation import train_test_split
else:
from sklearn.model_selection import train_test_split
infertdf = infert.as_df()
infertdf["isCase"] = infertdf.case == 1
data_train, data_test, y_train, y_test = train_test_split(infertdf, infertdf.isCase)
forest_model = rx_neural_network(
formula=" isCase ~ age + parity + education + spontaneous + induced ",
data=data_train)
# RuntimeError: The type (RxTextData) for file is not supported.
score_ds = rx_predict(forest_model, data=data_test,
extra_vars_to_write=["isCase", "Score"])
# Print the first five rows
print(rx_data_step(score_ds, number_rows_read=5))
출력:
Automatically adding a MinMax normalization transform, use 'norm=Warn' or 'norm=No' to turn this behavior off.
Beginning processing data.
Rows Read: 186, Read Time: 0, Transform Time: 0
Beginning processing data.
Beginning processing data.
Rows Read: 186, Read Time: 0, Transform Time: 0
Beginning processing data.
Beginning processing data.
Rows Read: 186, Read Time: 0, Transform Time: 0
Beginning processing data.
Using: AVX Math
***** Net definition *****
input Data [5];
hidden H [100] sigmoid { // Depth 1
from Data all;
}
output Result [1] sigmoid { // Depth 0
from H all;
}
***** End net definition *****
Input count: 5
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, 701 Weights...
Estimated Pre-training MeanError = 0.742343
Iter:1/100, MeanErr=0.680245(-8.37%), 119.87M WeightUpdates/sec
Iter:2/100, MeanErr=0.637843(-6.23%), 122.52M WeightUpdates/sec
Iter:3/100, MeanErr=0.635404(-0.38%), 122.24M WeightUpdates/sec
Iter:4/100, MeanErr=0.634980(-0.07%), 73.36M WeightUpdates/sec
Iter:5/100, MeanErr=0.635287(0.05%), 128.26M WeightUpdates/sec
Iter:6/100, MeanErr=0.634572(-0.11%), 131.05M WeightUpdates/sec
Iter:7/100, MeanErr=0.634827(0.04%), 124.27M WeightUpdates/sec
Iter:8/100, MeanErr=0.635359(0.08%), 123.69M WeightUpdates/sec
Iter:9/100, MeanErr=0.635244(-0.02%), 119.35M WeightUpdates/sec
Iter:10/100, MeanErr=0.634712(-0.08%), 127.80M WeightUpdates/sec
Iter:11/100, MeanErr=0.635105(0.06%), 122.69M WeightUpdates/sec
Iter:12/100, MeanErr=0.635226(0.02%), 98.61M WeightUpdates/sec
Iter:13/100, MeanErr=0.634977(-0.04%), 127.88M WeightUpdates/sec
Iter:14/100, MeanErr=0.634347(-0.10%), 123.25M WeightUpdates/sec
Iter:15/100, MeanErr=0.634891(0.09%), 124.27M WeightUpdates/sec
Iter:16/100, MeanErr=0.635116(0.04%), 123.06M WeightUpdates/sec
Iter:17/100, MeanErr=0.633770(-0.21%), 122.05M WeightUpdates/sec
Iter:18/100, MeanErr=0.634992(0.19%), 128.79M WeightUpdates/sec
Iter:19/100, MeanErr=0.634385(-0.10%), 122.95M WeightUpdates/sec
Iter:20/100, MeanErr=0.634752(0.06%), 127.14M WeightUpdates/sec
Iter:21/100, MeanErr=0.635043(0.05%), 123.44M WeightUpdates/sec
Iter:22/100, MeanErr=0.634845(-0.03%), 121.81M WeightUpdates/sec
Iter:23/100, MeanErr=0.634850(0.00%), 125.11M WeightUpdates/sec
Iter:24/100, MeanErr=0.634617(-0.04%), 122.18M WeightUpdates/sec
Iter:25/100, MeanErr=0.634675(0.01%), 125.69M WeightUpdates/sec
Iter:26/100, MeanErr=0.634911(0.04%), 122.44M WeightUpdates/sec
Iter:27/100, MeanErr=0.634311(-0.09%), 121.90M WeightUpdates/sec
Iter:28/100, MeanErr=0.634798(0.08%), 123.54M WeightUpdates/sec
Iter:29/100, MeanErr=0.634674(-0.02%), 127.53M WeightUpdates/sec
Iter:30/100, MeanErr=0.634546(-0.02%), 100.96M WeightUpdates/sec
Iter:31/100, MeanErr=0.634859(0.05%), 124.40M WeightUpdates/sec
Iter:32/100, MeanErr=0.634747(-0.02%), 128.21M WeightUpdates/sec
Iter:33/100, MeanErr=0.634842(0.02%), 125.82M WeightUpdates/sec
Iter:34/100, MeanErr=0.634703(-0.02%), 77.48M WeightUpdates/sec
Iter:35/100, MeanErr=0.634804(0.02%), 122.21M WeightUpdates/sec
Iter:36/100, MeanErr=0.634690(-0.02%), 112.48M WeightUpdates/sec
Iter:37/100, MeanErr=0.634654(-0.01%), 119.18M WeightUpdates/sec
Iter:38/100, MeanErr=0.634885(0.04%), 137.19M WeightUpdates/sec
Iter:39/100, MeanErr=0.634723(-0.03%), 113.80M WeightUpdates/sec
Iter:40/100, MeanErr=0.634714(0.00%), 127.50M WeightUpdates/sec
Iter:41/100, MeanErr=0.634794(0.01%), 129.54M WeightUpdates/sec
Iter:42/100, MeanErr=0.633835(-0.15%), 133.05M WeightUpdates/sec
Iter:43/100, MeanErr=0.634401(0.09%), 128.95M WeightUpdates/sec
Iter:44/100, MeanErr=0.634575(0.03%), 123.42M WeightUpdates/sec
Iter:45/100, MeanErr=0.634673(0.02%), 123.78M WeightUpdates/sec
Iter:46/100, MeanErr=0.634692(0.00%), 119.04M WeightUpdates/sec
Iter:47/100, MeanErr=0.634476(-0.03%), 122.95M WeightUpdates/sec
Iter:48/100, MeanErr=0.634583(0.02%), 97.87M WeightUpdates/sec
Iter:49/100, MeanErr=0.634706(0.02%), 121.41M WeightUpdates/sec
Iter:50/100, MeanErr=0.634564(-0.02%), 120.58M WeightUpdates/sec
Iter:51/100, MeanErr=0.634118(-0.07%), 120.17M WeightUpdates/sec
Iter:52/100, MeanErr=0.634699(0.09%), 127.27M WeightUpdates/sec
Iter:53/100, MeanErr=0.634123(-0.09%), 110.51M WeightUpdates/sec
Iter:54/100, MeanErr=0.634390(0.04%), 123.74M WeightUpdates/sec
Iter:55/100, MeanErr=0.634461(0.01%), 113.66M WeightUpdates/sec
Iter:56/100, MeanErr=0.634415(-0.01%), 118.61M WeightUpdates/sec
Iter:57/100, MeanErr=0.634453(0.01%), 114.99M WeightUpdates/sec
Iter:58/100, MeanErr=0.634478(0.00%), 104.53M WeightUpdates/sec
Iter:59/100, MeanErr=0.634010(-0.07%), 124.62M WeightUpdates/sec
Iter:60/100, MeanErr=0.633901(-0.02%), 118.93M WeightUpdates/sec
Iter:61/100, MeanErr=0.634088(0.03%), 40.46M WeightUpdates/sec
Iter:62/100, MeanErr=0.634046(-0.01%), 94.65M WeightUpdates/sec
Iter:63/100, MeanErr=0.634233(0.03%), 27.18M WeightUpdates/sec
Iter:64/100, MeanErr=0.634596(0.06%), 123.94M WeightUpdates/sec
Iter:65/100, MeanErr=0.634185(-0.06%), 125.01M WeightUpdates/sec
Iter:66/100, MeanErr=0.634469(0.04%), 119.41M WeightUpdates/sec
Iter:67/100, MeanErr=0.634333(-0.02%), 124.11M WeightUpdates/sec
Iter:68/100, MeanErr=0.634203(-0.02%), 112.68M WeightUpdates/sec
Iter:69/100, MeanErr=0.633854(-0.05%), 118.62M WeightUpdates/sec
Iter:70/100, MeanErr=0.634319(0.07%), 123.59M WeightUpdates/sec
Iter:71/100, MeanErr=0.634423(0.02%), 122.51M WeightUpdates/sec
Iter:72/100, MeanErr=0.634388(-0.01%), 126.15M WeightUpdates/sec
Iter:73/100, MeanErr=0.634230(-0.02%), 126.51M WeightUpdates/sec
Iter:74/100, MeanErr=0.634011(-0.03%), 128.32M WeightUpdates/sec
Iter:75/100, MeanErr=0.634294(0.04%), 127.48M WeightUpdates/sec
Iter:76/100, MeanErr=0.634372(0.01%), 123.51M WeightUpdates/sec
Iter:77/100, MeanErr=0.632020(-0.37%), 122.12M WeightUpdates/sec
Iter:78/100, MeanErr=0.633770(0.28%), 119.55M WeightUpdates/sec
Iter:79/100, MeanErr=0.633504(-0.04%), 124.21M WeightUpdates/sec
Iter:80/100, MeanErr=0.634154(0.10%), 125.94M WeightUpdates/sec
Iter:81/100, MeanErr=0.633491(-0.10%), 120.83M WeightUpdates/sec
Iter:82/100, MeanErr=0.634212(0.11%), 128.60M WeightUpdates/sec
Iter:83/100, MeanErr=0.634138(-0.01%), 73.58M WeightUpdates/sec
Iter:84/100, MeanErr=0.634244(0.02%), 124.08M WeightUpdates/sec
Iter:85/100, MeanErr=0.634065(-0.03%), 96.43M WeightUpdates/sec
Iter:86/100, MeanErr=0.634174(0.02%), 124.28M WeightUpdates/sec
Iter:87/100, MeanErr=0.633966(-0.03%), 125.24M WeightUpdates/sec
Iter:88/100, MeanErr=0.633989(0.00%), 130.31M WeightUpdates/sec
Iter:89/100, MeanErr=0.633767(-0.04%), 115.73M WeightUpdates/sec
Iter:90/100, MeanErr=0.633831(0.01%), 122.81M WeightUpdates/sec
Iter:91/100, MeanErr=0.633219(-0.10%), 114.91M WeightUpdates/sec
Iter:92/100, MeanErr=0.633589(0.06%), 93.29M WeightUpdates/sec
Iter:93/100, MeanErr=0.634086(0.08%), 123.31M WeightUpdates/sec
Iter:94/100, MeanErr=0.634075(0.00%), 120.99M WeightUpdates/sec
Iter:95/100, MeanErr=0.634071(0.00%), 122.49M WeightUpdates/sec
Iter:96/100, MeanErr=0.633523(-0.09%), 116.48M WeightUpdates/sec
Iter:97/100, MeanErr=0.634103(0.09%), 128.85M WeightUpdates/sec
Iter:98/100, MeanErr=0.633836(-0.04%), 123.87M WeightUpdates/sec
Iter:99/100, MeanErr=0.633772(-0.01%), 128.17M WeightUpdates/sec
Iter:100/100, MeanErr=0.633684(-0.01%), 123.65M WeightUpdates/sec
Done!
Estimated Post-training MeanError = 0.631268
___________________________________________________________________
Not training a calibrator because it is not needed.
Elapsed time: 00:00:00.2454094
Elapsed time: 00:00:00.0082325
Beginning processing data.
Rows Read: 62, Read Time: 0.001, Transform Time: 0
Beginning processing data.
Elapsed time: 00:00:00.0297006
Finished writing 62 rows.
Writing completed.
Rows Read: 5, Total Rows Processed: 5, Total Chunk Time: 0.001 seconds
isCase PredictedLabel Score Probability
0 True False -0.689636 0.334114
1 True False -0.710219 0.329551
2 True False -0.712912 0.328956
3 False False -0.700765 0.331643
4 True False -0.689783 0.334081
MultiClass 분류 예제
'''
MultiClass Classification.
'''
import numpy
import pandas
from microsoftml import rx_neural_network, rx_predict
from revoscalepy.etl.RxDataStep import rx_data_step
from microsoftml.datasets.datasets import get_dataset
iris = get_dataset("iris")
import sklearn
if sklearn.__version__ < "0.18":
from sklearn.cross_validation import train_test_split
else:
from sklearn.model_selection import train_test_split
irisdf = iris.as_df()
irisdf["Species"] = irisdf["Species"].astype("category")
data_train, data_test, y_train, y_test = train_test_split(irisdf, irisdf.Species)
model = rx_neural_network(
formula=" Species ~ Sepal_Length + Sepal_Width + Petal_Length + Petal_Width ",
method="multiClass",
data=data_train)
# RuntimeError: The type (RxTextData) for file is not supported.
score_ds = rx_predict(model, data=data_test,
extra_vars_to_write=["Species", "Score"])
# Print the first five rows
print(rx_data_step(score_ds, number_rows_read=5))
출력:
Automatically adding a MinMax normalization transform, use 'norm=Warn' or 'norm=No' to turn this behavior off.
Beginning processing data.
Rows Read: 112, Read Time: 0.001, Transform Time: 0
Beginning processing data.
Beginning processing data.
Rows Read: 112, Read Time: 0, Transform Time: 0
Beginning processing data.
Beginning processing data.
Rows Read: 112, Read Time: 0, Transform Time: 0
Beginning processing data.
Using: AVX Math
***** Net definition *****
input Data [4];
hidden H [100] sigmoid { // Depth 1
from Data all;
}
output Result [3] softmax { // Depth 0
from H all;
}
***** End net definition *****
Input count: 4
Output count: 3
Output Function: SoftMax
Loss Function: LogLoss
PreTrainer: NoPreTrainer
___________________________________________________________________
Starting training...
Learning rate: 0.001000
Momentum: 0.000000
InitWtsDiameter: 0.100000
___________________________________________________________________
Initializing 1 Hidden Layers, 803 Weights...
Estimated Pre-training MeanError = 1.949606
Iter:1/100, MeanErr=1.937924(-0.60%), 98.43M WeightUpdates/sec
Iter:2/100, MeanErr=1.921153(-0.87%), 96.21M WeightUpdates/sec
Iter:3/100, MeanErr=1.920000(-0.06%), 95.55M WeightUpdates/sec
Iter:4/100, MeanErr=1.917267(-0.14%), 81.25M WeightUpdates/sec
Iter:5/100, MeanErr=1.917611(0.02%), 102.44M WeightUpdates/sec
Iter:6/100, MeanErr=1.918476(0.05%), 106.16M WeightUpdates/sec
Iter:7/100, MeanErr=1.916096(-0.12%), 97.85M WeightUpdates/sec
Iter:8/100, MeanErr=1.919486(0.18%), 77.99M WeightUpdates/sec
Iter:9/100, MeanErr=1.916452(-0.16%), 95.67M WeightUpdates/sec
Iter:10/100, MeanErr=1.916024(-0.02%), 102.06M WeightUpdates/sec
Iter:11/100, MeanErr=1.917155(0.06%), 99.21M WeightUpdates/sec
Iter:12/100, MeanErr=1.918543(0.07%), 99.25M WeightUpdates/sec
Iter:13/100, MeanErr=1.919120(0.03%), 85.38M WeightUpdates/sec
Iter:14/100, MeanErr=1.917713(-0.07%), 103.00M WeightUpdates/sec
Iter:15/100, MeanErr=1.917675(0.00%), 98.70M WeightUpdates/sec
Iter:16/100, MeanErr=1.917982(0.02%), 99.10M WeightUpdates/sec
Iter:17/100, MeanErr=1.916254(-0.09%), 103.41M WeightUpdates/sec
Iter:18/100, MeanErr=1.915691(-0.03%), 102.00M WeightUpdates/sec
Iter:19/100, MeanErr=1.914844(-0.04%), 86.64M WeightUpdates/sec
Iter:20/100, MeanErr=1.919268(0.23%), 94.68M WeightUpdates/sec
Iter:21/100, MeanErr=1.918748(-0.03%), 108.11M WeightUpdates/sec
Iter:22/100, MeanErr=1.917997(-0.04%), 96.33M WeightUpdates/sec
Iter:23/100, MeanErr=1.914987(-0.16%), 82.84M WeightUpdates/sec
Iter:24/100, MeanErr=1.916550(0.08%), 99.70M WeightUpdates/sec
Iter:25/100, MeanErr=1.915401(-0.06%), 96.69M WeightUpdates/sec
Iter:26/100, MeanErr=1.916092(0.04%), 101.62M WeightUpdates/sec
Iter:27/100, MeanErr=1.916381(0.02%), 98.81M WeightUpdates/sec
Iter:28/100, MeanErr=1.917414(0.05%), 102.29M WeightUpdates/sec
Iter:29/100, MeanErr=1.917316(-0.01%), 100.17M WeightUpdates/sec
Iter:30/100, MeanErr=1.916507(-0.04%), 82.09M WeightUpdates/sec
Iter:31/100, MeanErr=1.915786(-0.04%), 98.33M WeightUpdates/sec
Iter:32/100, MeanErr=1.917581(0.09%), 101.70M WeightUpdates/sec
Iter:33/100, MeanErr=1.913680(-0.20%), 79.94M WeightUpdates/sec
Iter:34/100, MeanErr=1.917264(0.19%), 102.54M WeightUpdates/sec
Iter:35/100, MeanErr=1.917377(0.01%), 100.67M WeightUpdates/sec
Iter:36/100, MeanErr=1.912060(-0.28%), 70.37M WeightUpdates/sec
Iter:37/100, MeanErr=1.917009(0.26%), 80.80M WeightUpdates/sec
Iter:38/100, MeanErr=1.916216(-0.04%), 94.56M WeightUpdates/sec
Iter:39/100, MeanErr=1.916362(0.01%), 28.22M WeightUpdates/sec
Iter:40/100, MeanErr=1.910658(-0.30%), 100.87M WeightUpdates/sec
Iter:41/100, MeanErr=1.916375(0.30%), 85.99M WeightUpdates/sec
Iter:42/100, MeanErr=1.916257(-0.01%), 102.06M WeightUpdates/sec
Iter:43/100, MeanErr=1.914505(-0.09%), 99.86M WeightUpdates/sec
Iter:44/100, MeanErr=1.914638(0.01%), 103.11M WeightUpdates/sec
Iter:45/100, MeanErr=1.915141(0.03%), 107.62M WeightUpdates/sec
Iter:46/100, MeanErr=1.915119(0.00%), 99.65M WeightUpdates/sec
Iter:47/100, MeanErr=1.915379(0.01%), 107.03M WeightUpdates/sec
Iter:48/100, MeanErr=1.912565(-0.15%), 104.78M WeightUpdates/sec
Iter:49/100, MeanErr=1.915466(0.15%), 110.43M WeightUpdates/sec
Iter:50/100, MeanErr=1.914038(-0.07%), 98.44M WeightUpdates/sec
Iter:51/100, MeanErr=1.915015(0.05%), 96.28M WeightUpdates/sec
Iter:52/100, MeanErr=1.913771(-0.06%), 89.27M WeightUpdates/sec
Iter:53/100, MeanErr=1.911621(-0.11%), 72.67M WeightUpdates/sec
Iter:54/100, MeanErr=1.914969(0.18%), 111.17M WeightUpdates/sec
Iter:55/100, MeanErr=1.913894(-0.06%), 98.68M WeightUpdates/sec
Iter:56/100, MeanErr=1.914871(0.05%), 95.41M WeightUpdates/sec
Iter:57/100, MeanErr=1.912898(-0.10%), 80.72M WeightUpdates/sec
Iter:58/100, MeanErr=1.913334(0.02%), 103.71M WeightUpdates/sec
Iter:59/100, MeanErr=1.913362(0.00%), 99.57M WeightUpdates/sec
Iter:60/100, MeanErr=1.913915(0.03%), 106.21M WeightUpdates/sec
Iter:61/100, MeanErr=1.913310(-0.03%), 112.27M WeightUpdates/sec
Iter:62/100, MeanErr=1.913395(0.00%), 50.86M WeightUpdates/sec
Iter:63/100, MeanErr=1.912814(-0.03%), 58.91M WeightUpdates/sec
Iter:64/100, MeanErr=1.911468(-0.07%), 72.06M WeightUpdates/sec
Iter:65/100, MeanErr=1.912313(0.04%), 86.34M WeightUpdates/sec
Iter:66/100, MeanErr=1.913320(0.05%), 114.39M WeightUpdates/sec
Iter:67/100, MeanErr=1.912914(-0.02%), 105.97M WeightUpdates/sec
Iter:68/100, MeanErr=1.909881(-0.16%), 105.73M WeightUpdates/sec
Iter:69/100, MeanErr=1.911649(0.09%), 105.23M WeightUpdates/sec
Iter:70/100, MeanErr=1.911192(-0.02%), 110.24M WeightUpdates/sec
Iter:71/100, MeanErr=1.912480(0.07%), 106.86M WeightUpdates/sec
Iter:72/100, MeanErr=1.909881(-0.14%), 97.28M WeightUpdates/sec
Iter:73/100, MeanErr=1.911678(0.09%), 109.57M WeightUpdates/sec
Iter:74/100, MeanErr=1.911137(-0.03%), 91.01M WeightUpdates/sec
Iter:75/100, MeanErr=1.910706(-0.02%), 99.41M WeightUpdates/sec
Iter:76/100, MeanErr=1.910869(0.01%), 84.18M WeightUpdates/sec
Iter:77/100, MeanErr=1.911643(0.04%), 105.07M WeightUpdates/sec
Iter:78/100, MeanErr=1.911438(-0.01%), 110.12M WeightUpdates/sec
Iter:79/100, MeanErr=1.909590(-0.10%), 84.16M WeightUpdates/sec
Iter:80/100, MeanErr=1.911181(0.08%), 92.30M WeightUpdates/sec
Iter:81/100, MeanErr=1.910534(-0.03%), 110.60M WeightUpdates/sec
Iter:82/100, MeanErr=1.909340(-0.06%), 54.07M WeightUpdates/sec
Iter:83/100, MeanErr=1.908275(-0.06%), 104.08M WeightUpdates/sec
Iter:84/100, MeanErr=1.910364(0.11%), 107.19M WeightUpdates/sec
Iter:85/100, MeanErr=1.910286(0.00%), 102.55M WeightUpdates/sec
Iter:86/100, MeanErr=1.909155(-0.06%), 79.72M WeightUpdates/sec
Iter:87/100, MeanErr=1.909384(0.01%), 102.37M WeightUpdates/sec
Iter:88/100, MeanErr=1.907751(-0.09%), 105.48M WeightUpdates/sec
Iter:89/100, MeanErr=1.910164(0.13%), 102.53M WeightUpdates/sec
Iter:90/100, MeanErr=1.907935(-0.12%), 105.03M WeightUpdates/sec
Iter:91/100, MeanErr=1.909510(0.08%), 99.97M WeightUpdates/sec
Iter:92/100, MeanErr=1.907405(-0.11%), 100.03M WeightUpdates/sec
Iter:93/100, MeanErr=1.905757(-0.09%), 113.21M WeightUpdates/sec
Iter:94/100, MeanErr=1.909167(0.18%), 107.86M WeightUpdates/sec
Iter:95/100, MeanErr=1.907593(-0.08%), 106.09M WeightUpdates/sec
Iter:96/100, MeanErr=1.908358(0.04%), 111.25M WeightUpdates/sec
Iter:97/100, MeanErr=1.906484(-0.10%), 95.81M WeightUpdates/sec
Iter:98/100, MeanErr=1.908239(0.09%), 105.89M WeightUpdates/sec
Iter:99/100, MeanErr=1.908508(0.01%), 103.05M WeightUpdates/sec
Iter:100/100, MeanErr=1.904747(-0.20%), 106.81M WeightUpdates/sec
Done!
Estimated Post-training MeanError = 1.896338
___________________________________________________________________
Not training a calibrator because it is not needed.
Elapsed time: 00:00:00.1620840
Elapsed time: 00:00:00.0096627
Beginning processing data.
Rows Read: 38, Read Time: 0, Transform Time: 0
Beginning processing data.
Elapsed time: 00:00:00.0312987
Finished writing 38 rows.
Writing completed.
Rows Read: 5, Total Rows Processed: 5, Total Chunk Time: Less than .001 seconds
Species Score.0 Score.1 Score.2
0 versicolor 0.350161 0.339557 0.310282
1 setosa 0.358506 0.336593 0.304901
2 virginica 0.346957 0.340573 0.312470
3 virginica 0.346685 0.340748 0.312567
4 virginica 0.348469 0.340113 0.311417
회귀 예제
'''
Regression.
'''
import numpy
import pandas
from microsoftml import rx_neural_network, rx_predict
from revoscalepy.etl.RxDataStep import rx_data_step
from microsoftml.datasets.datasets import get_dataset
attitude = get_dataset("attitude")
import sklearn
if sklearn.__version__ < "0.18":
from sklearn.cross_validation import train_test_split
else:
from sklearn.model_selection import train_test_split
attitudedf = attitude.as_df()
data_train, data_test = train_test_split(attitudedf)
model = rx_neural_network(
formula="rating ~ complaints + privileges + learning + raises + critical + advance",
method="regression",
data=data_train)
# RuntimeError: The type (RxTextData) for file is not supported.
score_ds = rx_predict(model, data=data_test,
extra_vars_to_write=["rating"])
# Print the first five rows
print(rx_data_step(score_ds, number_rows_read=5))
출력:
Automatically adding a MinMax normalization transform, use 'norm=Warn' or 'norm=No' to turn this behavior off.
Beginning processing data.
Rows Read: 22, Read Time: 0, Transform Time: 0
Beginning processing data.
Beginning processing data.
Rows Read: 22, Read Time: 0.001, Transform Time: 0
Beginning processing data.
Beginning processing data.
Rows Read: 22, Read Time: 0, Transform Time: 0
Beginning processing data.
Using: AVX Math
***** Net definition *****
input Data [6];
hidden H [100] sigmoid { // Depth 1
from Data all;
}
output Result [1] linear { // Depth 0
from H all;
}
***** End net definition *****
Input count: 6
Output count: 1
Output Function: Linear
Loss Function: SquaredLoss
PreTrainer: NoPreTrainer
___________________________________________________________________
Starting training...
Learning rate: 0.001000
Momentum: 0.000000
InitWtsDiameter: 0.100000
___________________________________________________________________
Initializing 1 Hidden Layers, 801 Weights...
Estimated Pre-training MeanError = 4458.793673
Iter:1/100, MeanErr=1624.747024(-63.56%), 27.30M WeightUpdates/sec
Iter:2/100, MeanErr=139.267390(-91.43%), 30.50M WeightUpdates/sec
Iter:3/100, MeanErr=116.382316(-16.43%), 29.16M WeightUpdates/sec
Iter:4/100, MeanErr=114.947244(-1.23%), 32.06M WeightUpdates/sec
Iter:5/100, MeanErr=112.886818(-1.79%), 32.96M WeightUpdates/sec
Iter:6/100, MeanErr=112.406547(-0.43%), 30.29M WeightUpdates/sec
Iter:7/100, MeanErr=110.502757(-1.69%), 30.92M WeightUpdates/sec
Iter:8/100, MeanErr=111.499645(0.90%), 31.20M WeightUpdates/sec
Iter:9/100, MeanErr=111.895816(0.36%), 32.46M WeightUpdates/sec
Iter:10/100, MeanErr=110.171443(-1.54%), 34.61M WeightUpdates/sec
Iter:11/100, MeanErr=106.975524(-2.90%), 22.14M WeightUpdates/sec
Iter:12/100, MeanErr=107.708220(0.68%), 7.73M WeightUpdates/sec
Iter:13/100, MeanErr=105.345097(-2.19%), 28.99M WeightUpdates/sec
Iter:14/100, MeanErr=109.937833(4.36%), 31.04M WeightUpdates/sec
Iter:15/100, MeanErr=106.672340(-2.97%), 30.04M WeightUpdates/sec
Iter:16/100, MeanErr=108.474555(1.69%), 32.41M WeightUpdates/sec
Iter:17/100, MeanErr=109.449054(0.90%), 31.60M WeightUpdates/sec
Iter:18/100, MeanErr=105.911830(-3.23%), 34.05M WeightUpdates/sec
Iter:19/100, MeanErr=106.045172(0.13%), 33.80M WeightUpdates/sec
Iter:20/100, MeanErr=108.360427(2.18%), 33.60M WeightUpdates/sec
Iter:21/100, MeanErr=106.506436(-1.71%), 33.77M WeightUpdates/sec
Iter:22/100, MeanErr=99.167335(-6.89%), 32.26M WeightUpdates/sec
Iter:23/100, MeanErr=108.115797(9.02%), 25.86M WeightUpdates/sec
Iter:24/100, MeanErr=106.292283(-1.69%), 31.03M WeightUpdates/sec
Iter:25/100, MeanErr=99.397875(-6.49%), 31.33M WeightUpdates/sec
Iter:26/100, MeanErr=104.805299(5.44%), 31.57M WeightUpdates/sec
Iter:27/100, MeanErr=101.385085(-3.26%), 22.92M WeightUpdates/sec
Iter:28/100, MeanErr=100.064656(-1.30%), 35.01M WeightUpdates/sec
Iter:29/100, MeanErr=100.519013(0.45%), 32.74M WeightUpdates/sec
Iter:30/100, MeanErr=99.273143(-1.24%), 35.12M WeightUpdates/sec
Iter:31/100, MeanErr=100.465649(1.20%), 33.68M WeightUpdates/sec
Iter:32/100, MeanErr=102.402320(1.93%), 33.79M WeightUpdates/sec
Iter:33/100, MeanErr=97.517196(-4.77%), 32.32M WeightUpdates/sec
Iter:34/100, MeanErr=102.597511(5.21%), 32.46M WeightUpdates/sec
Iter:35/100, MeanErr=96.187788(-6.25%), 32.32M WeightUpdates/sec
Iter:36/100, MeanErr=101.533507(5.56%), 21.44M WeightUpdates/sec
Iter:37/100, MeanErr=99.339624(-2.16%), 21.53M WeightUpdates/sec
Iter:38/100, MeanErr=98.049306(-1.30%), 15.27M WeightUpdates/sec
Iter:39/100, MeanErr=97.508282(-0.55%), 23.21M WeightUpdates/sec
Iter:40/100, MeanErr=99.894288(2.45%), 27.94M WeightUpdates/sec
Iter:41/100, MeanErr=95.190566(-4.71%), 32.47M WeightUpdates/sec
Iter:42/100, MeanErr=91.234977(-4.16%), 31.29M WeightUpdates/sec
Iter:43/100, MeanErr=98.824414(8.32%), 32.35M WeightUpdates/sec
Iter:44/100, MeanErr=96.759533(-2.09%), 22.37M WeightUpdates/sec
Iter:45/100, MeanErr=95.275106(-1.53%), 32.09M WeightUpdates/sec
Iter:46/100, MeanErr=95.749031(0.50%), 26.49M WeightUpdates/sec
Iter:47/100, MeanErr=96.267879(0.54%), 31.81M WeightUpdates/sec
Iter:48/100, MeanErr=97.383752(1.16%), 31.01M WeightUpdates/sec
Iter:49/100, MeanErr=96.605199(-0.80%), 32.05M WeightUpdates/sec
Iter:50/100, MeanErr=96.927400(0.33%), 32.42M WeightUpdates/sec
Iter:51/100, MeanErr=96.288491(-0.66%), 28.89M WeightUpdates/sec
Iter:52/100, MeanErr=92.751171(-3.67%), 33.68M WeightUpdates/sec
Iter:53/100, MeanErr=88.655001(-4.42%), 34.53M WeightUpdates/sec
Iter:54/100, MeanErr=90.923513(2.56%), 32.00M WeightUpdates/sec
Iter:55/100, MeanErr=91.627261(0.77%), 25.74M WeightUpdates/sec
Iter:56/100, MeanErr=91.132907(-0.54%), 30.00M WeightUpdates/sec
Iter:57/100, MeanErr=95.294092(4.57%), 33.13M WeightUpdates/sec
Iter:58/100, MeanErr=90.219024(-5.33%), 31.70M WeightUpdates/sec
Iter:59/100, MeanErr=92.727605(2.78%), 30.71M WeightUpdates/sec
Iter:60/100, MeanErr=86.910488(-6.27%), 33.07M WeightUpdates/sec
Iter:61/100, MeanErr=92.350984(6.26%), 32.46M WeightUpdates/sec
Iter:62/100, MeanErr=93.208298(0.93%), 31.08M WeightUpdates/sec
Iter:63/100, MeanErr=90.784723(-2.60%), 21.19M WeightUpdates/sec
Iter:64/100, MeanErr=88.685225(-2.31%), 33.17M WeightUpdates/sec
Iter:65/100, MeanErr=91.668555(3.36%), 30.65M WeightUpdates/sec
Iter:66/100, MeanErr=82.607568(-9.88%), 29.72M WeightUpdates/sec
Iter:67/100, MeanErr=88.787842(7.48%), 32.98M WeightUpdates/sec
Iter:68/100, MeanErr=88.793186(0.01%), 34.67M WeightUpdates/sec
Iter:69/100, MeanErr=88.918795(0.14%), 14.09M WeightUpdates/sec
Iter:70/100, MeanErr=87.121434(-2.02%), 33.02M WeightUpdates/sec
Iter:71/100, MeanErr=86.865602(-0.29%), 34.87M WeightUpdates/sec
Iter:72/100, MeanErr=87.261979(0.46%), 32.34M WeightUpdates/sec
Iter:73/100, MeanErr=87.812460(0.63%), 31.35M WeightUpdates/sec
Iter:74/100, MeanErr=87.818462(0.01%), 32.54M WeightUpdates/sec
Iter:75/100, MeanErr=87.085672(-0.83%), 34.80M WeightUpdates/sec
Iter:76/100, MeanErr=85.773668(-1.51%), 35.39M WeightUpdates/sec
Iter:77/100, MeanErr=85.338703(-0.51%), 34.59M WeightUpdates/sec
Iter:78/100, MeanErr=79.370105(-6.99%), 30.14M WeightUpdates/sec
Iter:79/100, MeanErr=83.026209(4.61%), 32.32M WeightUpdates/sec
Iter:80/100, MeanErr=89.776417(8.13%), 33.14M WeightUpdates/sec
Iter:81/100, MeanErr=85.447100(-4.82%), 32.32M WeightUpdates/sec
Iter:82/100, MeanErr=83.991969(-1.70%), 22.12M WeightUpdates/sec
Iter:83/100, MeanErr=85.065064(1.28%), 30.41M WeightUpdates/sec
Iter:84/100, MeanErr=83.762008(-1.53%), 31.29M WeightUpdates/sec
Iter:85/100, MeanErr=84.217726(0.54%), 34.92M WeightUpdates/sec
Iter:86/100, MeanErr=82.395181(-2.16%), 34.26M WeightUpdates/sec
Iter:87/100, MeanErr=82.979145(0.71%), 22.87M WeightUpdates/sec
Iter:88/100, MeanErr=83.656685(0.82%), 28.51M WeightUpdates/sec
Iter:89/100, MeanErr=81.132468(-3.02%), 32.43M WeightUpdates/sec
Iter:90/100, MeanErr=81.311106(0.22%), 30.91M WeightUpdates/sec
Iter:91/100, MeanErr=81.953897(0.79%), 31.98M WeightUpdates/sec
Iter:92/100, MeanErr=79.018074(-3.58%), 33.13M WeightUpdates/sec
Iter:93/100, MeanErr=78.220412(-1.01%), 31.47M WeightUpdates/sec
Iter:94/100, MeanErr=80.833884(3.34%), 25.16M WeightUpdates/sec
Iter:95/100, MeanErr=81.550135(0.89%), 32.64M WeightUpdates/sec
Iter:96/100, MeanErr=77.785628(-4.62%), 32.54M WeightUpdates/sec
Iter:97/100, MeanErr=76.438158(-1.73%), 34.34M WeightUpdates/sec
Iter:98/100, MeanErr=79.471621(3.97%), 33.12M WeightUpdates/sec
Iter:99/100, MeanErr=76.038475(-4.32%), 33.01M WeightUpdates/sec
Iter:100/100, MeanErr=75.349164(-0.91%), 32.68M WeightUpdates/sec
Done!
Estimated Post-training MeanError = 75.768932
___________________________________________________________________
Not training a calibrator because it is not needed.
Elapsed time: 00:00:00.1178557
Elapsed time: 00:00:00.0088299
Beginning processing data.
Rows Read: 8, Read Time: 0, Transform Time: 0
Beginning processing data.
Elapsed time: 00:00:00.0293893
Finished writing 8 rows.
Writing completed.
Rows Read: 5, Total Rows Processed: 5, Total Chunk Time: 0.001 seconds
rating Score
0 82.0 70.120613
1 64.0 66.344688
2 68.0 68.862373
3 58.0 68.241341
4 63.0 67.196869