pytorch/examples

example time_sequence_prediction does not produce expected output

Open

#247 建立於 2017年10月27日

在 GitHub 查看
 (5 留言) (1 反應) (0 負責人)Python (21,634 star) (9,429 fork)batch import
good first issue

描述

https://github.com/pytorch/examples/tree/master/time_sequence_prediction

After a few steps, the loss suddenly jumps up and stays high: loss: 0.00101216329969 loss: 0.195070432582

the pdf0 is OK, pdf1 is OK but from pdf2 the graph becomes an "L "

貢獻者指南

技術棧
pytorchpython
領域
machine learningai
議題類型
bug
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。
4
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。
half day
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。
mostly clear
前置要求
PythonPyTorchRNN/LSTMTime series prediction
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。
30
研究方向
Investigate the training loop in the time sequence prediction example (https://github.com/pytorch/examples/tree/master/time sequence prediction). The issue describes a sudden loss increase after a few steps and an 'L' shaped graph from pdf2 onward. Check for exploding gradients, learning rate issues, or incorrect sequence batching. Compare with well known implementations of LSTM for time series to identify potential bugs.