2013-10-30 102 views
0

我使用Encog,我使用SVM預測數據。我的訓練集值沒有進行歸一化處理,但它們最初在[-1,1]範圍內。我不明白爲什麼會出現這個問題。SVM爲每個輸入輸出相同的預測值...爲什麼是這樣?

我的訓練數據:

EURUSD_OPEN_CH,EURUSD_HIGH_CH,EURUSD_LOW_CH,EURUSD_CLOSE_CH,EURUSD_MACD,EURUSD_MACDS,EURUSD_STTDEV 
0.0134883819,0.0132838637,0.0135361889,0.0140344719,0.0023983892,0.0010403195,0.0054870487 
0.0001454143,0.0000969039,-0.0002216665,-0.0005261919,0.0035244907,0.0013168603,0.0070012526 
-0.0005261846,0.0006574986,0.0001593581,0.0009628839,0.0044774819,0.0017225556,0.0081131621 
0.0009282350,-0.0001867452,-0.0004156506,-0.0005882475,0.0051052958,0.0021969854,0.0088044648 
-0.0005605769,-0.0006641071,0.0001455382,0.0000069246,0.0055397905,0.0027231400,0.0092672117 
(...) 

我應該正常化這些價值?我認爲這不是一個問題,但是誰知道......我訓練SVM並且一切正常,但是當我評估SVM時,每個輸入的輸出都是相同的。如果需要的話,我可以附上代碼。

+0

「如果有必要,我可以附上代碼。」 - 是的,通常是一個不錯的選擇。 – AGS

回答

3

我是這樣一個noob ...規範化解決了這個問題。這些值太小而無法預測,所以我將整個CSV歸一化到了[0.1,0.9]範圍,並且它有所幫助。

+2

感謝您幫助另一個「noob」解決不同的問題。 –

+0

你是如何非規範化的? –

相關問題