'find_chains' not returning the longest chain in some datasets

[PrakritiAilavadi]

** 'find_chains' not returning the longest chain sometimes **
I've tested the function on datasets given on the official page of Time Series Chains (https://sites.google.com/site/timeserieschain/).
Performs beautifully on some, but fails on some others. In the chain calibration dataset given in the official page, on the dataset 'chain_test_1' the result is correct, but on the 3 other datasets, the output in best chain of the function is NULL, when I could clearly see through the total chains found that longest chain do exist.

To Reproduce
Data:

1. Go to https://sites.google.com/site/timeserieschain/
2. Under 'Chain Calibration Datasets', click on 'datasets'.

Code:

#each of subsequence length 127 
#each of about 6000-7000 data points

#chain test 1 - giving accurate results for best chain
#chain test 2,3,4 - giving best chain output as NULL.

chain_test_dataset <- read.delim("chain_test_2.txt", header=FALSE)
mp_chain_test <- tsmp(chain_test_dataset$V1, window_size = 127)
window_size_chaintest=127
chains_of_chain_test <- find_chains(mp_chain_test)
chains_of_chain_test
length_of_best= length(chains_of_chain_test$chain$best)
length_of_best

#Similarly for chain test 3 and 4

Screenshots

System used:

• OS: macOS Mojave
• R version: 1.0.143 – © 2009-2016
• 'tsmp' package version: 0.3.1

[franzbischoff]

Hello, It seems the Matrix Profile object is not correct either, as it doesn't say how many observations it has. Try to force data into a vertical matrix. I'll look into this if you can provide a sample of your data frame that reproduces the error.

…

-- Francisco Bischoff *"O mate está para o gaúcho como o chá para os ingleses, a coca para os bolivianos, o uísque para os escoceses e o café... para os brasileiros"-- Eduardo Bueno*

On Sat, Nov 3, 2018 at 6:22 PM PrakritiAilavadi ***@***.***> wrote: ** 'find_chains' not returning the longest chain sometimes ** I've tested the function on datasets given on the official page of Time Series Chains (https://sites.google.com/site/timeserieschain/). Performs beautifully on some, but fails on some others. In the chain calibration dataset given in the official page, on the dataset 'chain_test_1' the result is correct, but on the 3 other datasets, the output in best chain of the function is NULL, when I could clearly see through the total chains found that longest chain do exist. *To Reproduce* Data: 1. Go to https://sites.google.com/site/timeserieschain/ 2. Under 'Chain Calibration Datasets', click on 'datasets'. Code: `# each of subsequence length 127 each of about 6000-7000 data points #chain test 1 - giving accurate results for best chain chain test 2,3,4 - giving best chain output as NULL. chain_test_dataset <- read.delim("chain_test_2.txt", header=FALSE) mp_chain_test <- tsmp(chain_test_dataset$V1, window_size = 127) window_size_chaintest=127 chains_of_chain_test <- find_chains(mp_chain_test) chains_of_chain_test length_of_best= length(chains_of_chain_test$chain$best) length_of_best Similarly for chain test 3 and 4` *Screenshots* [image: screenshot 2018-11-03 at 11 35 18 pm] <https://user-images.githubusercontent.com/20098293/47955710-e6489a00-dfc1-11e8-946f-ff040ad0e8d2.png> [image: screenshot 2018-11-03 at 11 44 33 pm] <https://user-images.githubusercontent.com/20098293/47955749-9fa76f80-dfc2-11e8-806f-1a6063cae85c.png> *System used:* - OS: macOS Mojave - R version: 1.0.143 – © 2009-2016 - 'tsmp' package version: 0.3.1 — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <https://github.com/franzbischoff/tsmp/issues/33>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA8GEJx8IIzK0JVtreDFmoS0NGKHayTzks5urd7LgaJpZM4YM7pI> .

[PrakritiAilavadi]

The matrix profile object for chain_test_1 dataset also does not show the number of observations but is appropriately giving the best chain.

All datasets are already fed as a vertical matrix/column.

Sample datasets chain_test1,2,3,4 are attached.

Screenshot of chain_test_1 code and results:

Datasets:
chain_test_1.txt
chain_test_2.txt
chain_test_3.txt
chain_test_4.txt

All of the datasets have 6k-7k observations.

[franzbischoff]

Thanks. Have you tried the last version from github? I see you are not using the new progress bar.

…

-- Francisco Bischoff *"O mate está para o gaúcho como o chá para os ingleses, a coca para os bolivianos, o uísque para os escoceses e o café... para os brasileiros"-- Eduardo Bueno*

On Sun, Nov 4, 2018 at 6:44 PM PrakritiAilavadi ***@***.***> wrote: The matrix profile object for chain_test_1 dataset also does not show the number of observations but is appropriately giving the best chain. All datasets are already fed as a vertical matrix/column. Sample datasets chain_test1,2,3,4 are attached. Screenshot of chain_test_1 code and results: [image: screenshot 2018-11-05 at 12 12 11 am] <https://user-images.githubusercontent.com/20098293/47968403-7c96c180-e08f-11e8-9336-2c71843c900a.png> Datasets: chain_test_1.txt <https://github.com/franzbischoff/tsmp/files/2546147/chain_test_1.txt> chain_test_2.txt <https://github.com/franzbischoff/tsmp/files/2546148/chain_test_2.txt> chain_test_3.txt <https://github.com/franzbischoff/tsmp/files/2546149/chain_test_3.txt> chain_test_4.txt <https://github.com/franzbischoff/tsmp/files/2546150/chain_test_4.txt> All of the datasets have 6k-7k observations. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <https://github.com/franzbischoff/tsmp/issues/33#issuecomment-435694936>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA8GEJiM6TvCEP48dLWzbzTi3KNBjW3Vks5urzWQgaJpZM4YM7pI> .

[franzbischoff]

Bug confirmed, looking into it

[franzbischoff]

Bug is in line 55:
n <- mean(.mp$rmp[chain_set[[i]]])

one of the values is Inf, messing with the mean.