facebookarchive/caffe2

Accuracy requires INT32 while OneHot requires INT64

Open

#1,402 建立於 2017年10月30日

在 GitHub 查看
 (1 留言) (0 反應) (0 負責人)Shell (8,424 star) (1,954 fork)batch import
enhancementhelp wanted

描述

The operator Accuracy() takes parameter labels as type core.DataType.INT32. OneHot() takes parameter indices as type core.DataType.INT64. LabelCrossEntropy() takes label as type core.DataType.INT32. It's a bit waste of time to cast the label back and forth, hope there is alignment. Thx.

貢獻者指南

技術棧
pythoncpp
領域
machine learning
議題類型
feature
難度面向新貢獻者的預計實作難度，1 表示很小改動，5 表示專家級工作。
3
預計時間有經驗貢獻者完成調查、實作、測試並準備 pull request 的粗略時間範圍。
1-3 hours
活動狀態議題目前的可參與程度：新鮮、活躍、陳舊、阻塞或等待維護者輸入。
stale
清晰度議題是否清楚說明預期改動、驗收標準和下一步。
clear
前置要求
understanding of Caffe2 operator API
新手友善度1-100 的估計分數，表示該議題對首次貢獻者的友善程度。
50
研究方向
The Accuracy operator expects INT32 labels while OneHot expects INT64 indices, causing a type mismatch. To resolve, examine the source files for AccuracyOp, OneHotOp, and LabelCrossEntropyOp in the caffe2/operators directory to understand their label parameter types. Then propose a consistent type (e.g., change OneHot to use INT32) and update the corresponding tests and documentation. The single comment on the issue may provide additional context.
Accuracy requires INT32 while OneHot requires INT64 · facebookarchive/caffe2#1402 | Good First Issue