log.txt

Fri 10 Mar 2017
===============

d2:
"output/functions_systematic_ICD_CSI_MKL.csv"

Sun 12 Mar 2017
===============
d2: 37924.funcdebug
output/funcdebug
Simple graphs revealing the relationship between regularization, precision and Kendall.

Mon 13 Mar 2017
===============
d2: 37924.funcdebug
output/funcdebug
Add larger ranks to see whether best correlation persists at lambda=0.

d2: 34221.mklaren_functions_cso
Compute weights by norm approximation for both ICD and Mklaren.

Tue 21 Mar 2017
===============

[detached from 14942.mklaren]
Fitting lambda with cross-validation.

Wed 29 Mar 2017
===============
d3: [detached from 135022.mklaren]
Testing timing properties for up to 1e2 - 1e6 examples. (timing.py)

Wed 17 May 2017
===============
d:
    [detached from 13185.mklaren_snr] running snr.py
    [detached from 13185.mklaren_timing] running timing.py

Fri 19 May 2017
===============
d:
    [detached from 13185.mklaren_snr] running snr.py (Varying gamma, smaller rank, increase lambda range)


Sat 20 May 2017
===============

Repeat after NFS reset.
d:
    2775.mklaren_timing (05/20/2017 11:16:01 PM)    (Detached)
    2759.mklaren_snr    (05/20/2017 11:15:19 PM)    (Detached)


Mon 22 May 2017
===============

d:
   [detached from 2775.mklaren_timing]
   Repeating lambda-cv fitting for timing/2017-5-22/results_11.csv.


Mon 29 May 2017
===============
Running large-scale snr.py test for different levels of N, rank, gamma, sampling and noise.
[detached from 2759.mklaren_snr]
snr/2017-5-20/results_13.csv


Tue 30 May 2017
===============
# Added CSI and storage of distributions.
Running large-scale snr.py test for different levels of N, rank, gamma, sampling and noise.
[detached from 2759.mklaren_snr]
output/snr/2017-5-30/results_14.csv


Wed 31 May 2017
===============
Testing in 2D space.
[detached from 2759.mklaren_snr]
output/snr/2017-5-31/results_0.csv


Thu 15 Jun 2017
===============
Testing for FITC too.
output/snr/2017-6-15/results_0.csv

Thu 20 Jun 2017
===============
    Implemented cross-validation for all values of lambda.
        ../output/energy/2017-6-20/results_4.csv

Wed 21 Jun 2017
===============
Implemented periodic
    ../output/energy/2017-6-21/details_6/
and Matern kernel
    ../output/energy/2017-6-21/details_4/
and Exponential
    ../output/energy/2017-6-21/details_8/
scenarios.

modify
    export PYTHONPATH=../..
    export MKL_NUM_THREADS=7


Thu 22 Jul 2017
===============
Speed-up and enable FITC to use Matern32 kernel.

modify
    export PYTHONPATH=../..
    export MKL_NUM_THREADS=7

../output/energy/2017-6-22/details_0/ <- exponential [DONE]
../output/energy/2017-6-22/details_2/ <- periodic [NOK]
../output/energy/2017-6-22/details_4/ <- matern [SEGFAULT]


Mon 26 Jun 2017
===============
Running matern kernel for 10 signals in parallel on d:
../output/energy/2017-6-26/details_0/
-
../output/energy/2017-6-26/details_20/

Tue 27 Jun 2017
===============

# Merge results into one file
cd /Users/martin/Dev/mklaren/examples/output/energy/2017-6-26
head -n 1 results_0.csv > results.csv
grep -h "matern" results_*.csv >> results.csv

# Run Blitzer sentiment with L2 KRR (d)
blitzer_sentiment/2017-6-27/results_0.csv
blitzer_sentiment/2017-6-27/results_1.csv
blitzer_sentiment/2017-6-27/results_2.csv
blitzer_sentiment/2017-6-27/results_3.csv

# Repeating with custom kernel for each method (d3)
# Mklaren is not used here to save time. Merge all files into 1 in the end.
blitzer_sentiment/2017-6-27/results_4.csv
blitzer_sentiment/2017-6-27/results_5.csv
blitzer_sentiment/2017-6-27/results_6.csv
blitzer_sentiment/2017-6-27/results_7.csv


# Replacing tabulars in a directory with .tex tables
for f in *.tex ; do
    grep -v "table\|caption" $f > $f.temp
    mv $f.temp $f
done

# Sync with current results
rsync -ruvz blitzer_sentiment/ ~/PhD/writing/mkl-cholesky-lar/results/blitzer_sentiment/ --include=*.pdf --include=*.tex --exclude=*


Wed 28 Jun 2017
===============
Repeat blitzer_sentiment2 with correct L2-KRR method. The results here will be
equal to the uniform alignment, but at least there will be stored in one file.
2017-6-28/results_0.csv
2017-6-28/results_1.csv
2017-6-28/results_2.csv
2017-6-28/results_3.csv

Fri 30 Jun 2017
===============
Repeat Delve regression with varying number of kernels 5-100.

Sun 2 Jul 2017
==============
delve_regression2: Fix issues with CSI, add RFF_KMP, add two more datasets.
2017-7-2/results_0.csv
...
2017-7-2/results_7.csv

Mon 10 Jul 2017
===============
# Run delve on keel datasets
2017-7-10/results_0.csv
...
2017-7-10/results_28.csv

Mon 17 Jul 2017
===============
# Added explained variance to keel datasets
2017-7-17/results_0.csv
...
2017-7-17/results_28.csv


Tue 18 Jul 2017
===============
d;
Performance dependent on N.
(13750, 'ailerons')
(14998, 'pole')
(16599, 'elevators')
(20640, 'california')
(22784, 'house')
(40768, 'mv')

Wed 19 Jul 2017
===============
For 30k, 100k to illuminate the effect even more.
(Actual n is a little less, since not all datasets contain nuff samples)
python delve_regressionN.py ailerons
python delve_regressionN.py pole
python delve_regressionN.py elevators
python delve_regressionN.py california
python delve_regressionN.py house
python delve_regressionN.py mv

Thu 20 Jul 2017
===============
Testing on small geostats datasets.

Mon 31 Jul 2017
===============
# 145434.mklaren_fixed	(07/18/2017 10:28:28 AM)	(Detached)
# Repeating on n=3000 datasets
python delve_regression2.py boston
python delve_regression2.py abalone
python delve_regression2.py comp
python delve_regression2.py bank
python delve_regression2.py pumadyn
python delve_regression2.py kin
python delve_regression2.py ionosphere
python delve_regression2.py census

[detached from 179564.mklaren_string]
Running strig analysis on degenerate kernels.

# TODO: Synchronize output directories when done, by clearing up old stuff.


Tue 1 Aug 2017
==============
[detached from 179564.mklaren_string]
Running string analysis on degenerate kernels (N=500, L=500), slow.

Tue 2 Aug 2017
==============
[detached from 179564.mklaren_string]
Running strig analysis on a complete set kernels for K in (2, 5) (N=100, L=100).

[detached from 181198.mklaren_string_big]
Initial run of RNA context with spectrum 3-5 kernels.

[detached from 145434.mklaren_fixed]
Running with a complete set of kernels for K=4.

Thu 3 Aug 2017
==============
[detached from 145434.mklaren_fixed]
Running rnacontext with spectrum kernels 1-10.


Tue 8 Aug 2017
==============
d3: [detached from 2912.string_cv_val]
Repeating cv_val on d with plots.
Writing to ../output/string_lengthscales_cv_val/2017-8-8/results_0.csv


Wed 9 Aug 2017
==============
d3: [detached from 2579.rnacontext_kmer]
Running rnacontext again with spectrum kernels 1-10, rank 5 and generic function plot.
results 0-8.csv

Merge results into one file depending on kernel used:
    # Running periodic kernels for ICD, CSI, Nystrom, Mklaren
    (not implemented for FITC (no support in GPy, not our fault!), RFF_KMP)
    [detached from 4407.mklaren_periodic]

    # Additional tests for Matern due to a different hyperparameter range.
    [detached from 4663.mklaren_matern]


Thu 10 Aug 2017
==============

# Merge & cleanup after run:
    d3: 5564.string_cv	(08/10/2017 03:06:41 PM)	(Detached)
    Repeating string cv_lengthsales_val with RMSE.


# Merge & cleanup after run:
    d3: [detached from 2579.rnacontext_kmer]
    Running rnacontext again with spectrum kernels 1-10, rank 5 and generic function plot.
    Computing MSE.
    results 0-8.csv

    d3: [detached from 5809.rnacontext_kmer_alltest]
    Use all remaining points for testing.
    results 9-17.csv

d3: [detached from 6074.delve_eff_rank]
    Try delve back with comparable ranks for ICD2, CSI2, Nystrom2.
    The rows with each rank are now comparable.


Fri 11 Aug 2017
===============
d3: [detached from 32923.snr_timing]
Running the timing experiment for up to 1e6 points.
results_0.csv

Sat 12 Aug 2017
===============
d3: [detached from 6074.delve_eff_rank]
Repeating for CSI2, ICD2, Nystrom2 due to unexplicable errors.
Remember to merge on finish. OK

Sun 13 Aug 2017
===============
d3: [detached from 6074.delve_eff_rank]
Repeating everything with ranks 2-100 for all methods to use the time.


Mon 14 Aug 2017
===============
d3: 32923.snr_timing
Running timing for input dimensions 100-1.

d3: [detached from 5809.rnacontext_kmer_alltest]
Repeat rnacontext with ranks 10, 20


Tue 15 Aug 2017
===============
Increase lambda resolution and store chosen k-mer lengths.
[detached from 5809.rnacontext_kmer_alltest]


Wed 16 Aug 2017
===============
[detached from 123458.rnacontex_test]
Increase training set size to 2k.


Thu 17 Aug 2017
===============
123458.rnacontex_test	(08/16/2017 10:50:41 AM)	(Detached)
Changed the rnacontext dataset to "weak".
rnacontext/2017-8-17/results_*.csv  0-8
* WORKS
* tag: rnacontext_regr_canonical

Fri 18 Aug 2017
===============
[detached from 134075.rnacontext_spectrum]
Efficient pre-computation of spectrum kernels.
rnacontext/2017-8-18/results_*.csv 1-18
Increased training and validation sets to 3000.

134713.timing_test	(08/18/2017 04:47:08 PM)	(Detached)
Added CSI to timing test within a spearate process.


Sun 20 Aug 2017
===============
[detached from 145371.rnacotnext_spectrum]
rnacontext/2017-8-18/results_*.csv 1-18
Final try with training and validation sets to 1000.

Mon 21 Aug 2017
===============
[detached from 148946.snr_timing]
Re-run timings with fixed bugs on alignf and uniform.

Wed 23 Aug 2017
===============
[detached from 148946.snr_timing]
../output/snr/timings/2017-8-23/results_0.csv
Varying number of kernels.

../output/snr/timings/2017-8-23/results_1.csv
Repeating due to errors.

Fri 25 Aug 2017
===============
../output/snr/timings/2017-8-25/results_0.csv
Repeating due to errors.


----------------------------------- Re-run revision 2

Thu 12 Apr 2018
===============
d4: [detached from 348589.mklaren]
OUT=results/regr_mkl_cv_d4/
$PY regr_mkl_cv.py $OUT


Sun 29 Apr 2018
===============
# Repeating low-rank experiments including RFF non-stationary kernels.
d5:[detached from 307585.env]
OUT=./results-rffns/
/home/martins/Dev/py2/bin/python low_rank.py ANACALT ./results-rffns/ANACALT/


# Repeat experiments with time series
d5: [detached from 371696.B-env]
OUT=./results-matern
/home/martins/Dev/py2/bin/python energy.py T1 matern results-matern

Mon 30 Apr 2018
===============
# Repeat by fixing the lengthscales
d5: [detached from 371696.B-env]
OUT=./results-matern-scales
/home/martins/Dev/py2/bin/python energy.py T1 matern results-matern-scales


# Repeating low-rank experiments only with RFF non-stationary kernels.
d5:[detached from 307585.env]
OUT=./results-rff/
/home/martins/Dev/py2/bin/python low_rank.py ANACALT ./results-rff/ANACALT/


----------------------------------- Re-run revision-neuro

Fri 26 Oct 2018
===============
Re-run experiments for Neurocomputing revision.

Branch: revision-neuro
Continues from latest thesis updates.

# Add L2-KRR to MKL experiments
tag: l2krr_blitzer
d5: ./results-new

# Add Extrapolation experiment and Arima.
tag: arima_extrapolate
d5: ./results-extrapolate