keras-team / keras-docs-ja

Japanese translation of the Keras documentation.
https://keras.io/ja/
68 stars 67 forks source link

modified rescale description #110

Closed MasanoriTakaoka closed 6 years ago

MasanoriTakaoka commented 6 years ago

"rescaling factor" in original English text doesn't relate image size. Actually, rescale parameter affects image data like RGB data. rescale パラメータはサイズではなく、画素値を変換するものなのなので、誤解を招かないように「画像データ」としたほうが良いかと思います。

nzw0301 commented 6 years ago

Thank you for your PR. I think that 画像データの変更の係数 is an unclear description of this parameter. 標準化係数 may be better to explain the role of this parameter.

MasanoriTakaoka commented 6 years ago

Thanks for your comment. In my understanding, normalization 正規化 and standardization 標準化 are different meaning. In this case, I think normalization正規化 is preferable. So how about 画像データの正規化係数?

nzw0301 commented 6 years ago

Thank you for your quick comment. Yes, normalization and standardization are different. My comment was wrong. This parameter does not provide standardization.

At the same time, I disagree with you. This parameter is not normalization because the applied image vector's norm is not 1.. (Ref: scikit-learn's normalize)

Is 画素値のリスケーリング係数 better?

MasanoriTakaoka commented 6 years ago

Nice. I totally agree with your idea to avoid ambiguous of normalization or 正規化 in Japanese. Normalization has large meaning like using L1 norm, L2 norm, etc. I'll re-push. Anyway I'm happy to have a discussion with you. Thanks.

nzw0301 commented 6 years ago

Thank you for your PR and discussion!