search

AugurWorks / UI

Augurworks UI code
0 stars 0 forks source link

Date issue #90

Closed augurworks1 closed 10 years ago

augurworks1 commented 10 years ago

Strange results: I ran the normal LR today and we missed the prediction, which was a sizable change. So did a little digging. It looks like the dates a skewed over the weekend to the left looking back at the last few days. See https://docs.google.com/a/augurworks.com/spreadsheets/d/1FSuYBa1spmz_Ql4s01792RsoVyB84eUsZhWiqL8vpz0/edit#gid=1792976727 , where I tried to explain it.

The value of yesterday's daily prediction (2.84) now shows up at two days to the left with today's results. Interestingly, there are negative values in between yesterday's and today's prediction. Today the stock was negative. It seems like there is still something going on with the splining perhaps? If yesterday's 2.84 prediction was accurate, then it went negative as it should have (which is great). But, where did the following positive number come from for today?

082614 lr cms jpg

augurworks1 commented 10 years ago

See: https://docs.google.com/a/augurworks.com/spreadsheets/d/1FSuYBa1spmz_Ql4s01792RsoVyB84eUsZhWiqL8vpz0/edit#gid=1520043099

I went back through the beginning. For the most part, there are consistent discrepancies on Monday and Tuesday (between the "daily" values and the "history" values), which mostly straightens out by Wed through the remainder of the week. Thursday and Friday are almost always consistent. It seems related to the splining over the weekend??

augurworks1 commented 10 years ago

More data I believe is related. The circle was the LR predicted value for yesterday (.1). Between yesterday's LR prediction and today's prediction, there are three more data points; 6.40, 3.89 and -0.36. Today's LR prediction is -3.19. 3sept14 lr cms jpg

augurworks1 commented 10 years ago

Here's the same behavior using DTE with a little more detail. Yesterday's value was -2.60. Today's value is -3.38. There are two data points in between (5.48, 3.49 and -.59). The other must have had 3 data points to. 3sept14 dte lr jpg

augurworks1 commented 10 years ago

It actually skewed yesterday's data (.1 for CMS and -2.6 for DTE) back 3 days (representing Saturday's prediction). It looks like it added 3 days of splining data in between. It's 3 days because of the holiday. Last week it was 2 days. What doesn't make sense is the timing of when this happens where it seems to take until mid-week to correct. Strange. At least it's repeatable.

augurworks1 commented 10 years ago

Sorry one more. I reran (3 month so it's easier to see) yesterday's dates and the LR curve stayed the same as it is today. I saw this last week when I opened the ticket. It hard to put into words, so here's another pic.

cms 3mo jpg1

TheConnMan commented 10 years ago

I'm pretty sure this got fixed once we stopped jumping weekends.