Problem in nmf.LassoCV

[luca-dex]

When running nmf.LassoCV with the args reported in the vignettes the following error is shown:

> cv = nmf.LassoCV(x=patients,K=3:10)
Performing a grid search to estimate the best values of K and lambda with a total of 10 cross validation repetitions... 
starting worker pid=8516 on localhost:11120 at 11:06:40.616
starting worker pid=8530 on localhost:11120 at 11:06:40.754
starting worker pid=8539 on localhost:11120 at 11:06:40.890
starting worker pid=8548 on localhost:11120 at 11:06:41.026
starting worker pid=8557 on localhost:11120 at 11:06:41.165
starting worker pid=8566 on localhost:11120 at 11:06:41.303
starting worker pid=8575 on localhost:11120 at 11:06:41.443
Executing 7 processes via parallel... 
Starting cross validation with a total of 10 repetitions... 
Performing repetition 1 out of 10... 
Performing cross validation iteration 1 out of 5... 
Computing the initial values of beta by standard NMF... 
Timing stopped at: 0.002 0 0.002
Timing stopped at: 0.001 0 0.001
Timing stopped at: 0 0.001 0.001
Timing stopped at: 0.001 0 0.001
Timing stopped at: 0.001 0 0.001
Timing stopped at: 0.001 0 0.001
Timing stopped at: 0.001 0 0.001
Timing stopped at: 0.001 0 0.001
Error in (function (...)  : All the runs produced an error:
	-#1 [r=3] -> none of the packages are loaded [in call to 'path.package']
	-#2 [r=4] -> none of the packages are loaded [in call to 'path.package']
	-#3 [r=5] -> none of the packages are loaded [in call to 'path.package']
	-#4 [r=6] -> none of the packages are loaded [in call to 'path.package']
	-#5 [r=7] -> none of the packages are loaded [in call to 'path.package']
	-#6 [r=8] -> none of the packages are loaded [in call to 'path.package']
	-#7 [r=9] -> none of the packages are loaded [in call to 'path.package']
	-#8 [r=10] -> none of the packages are loaded [in call to 'path.package']

After loading the full NMF package this is the result:

> cv = nmf.LassoCV(x=patients,K=3:10)
Performing a grid search to estimate the best values of K and lambda with a total of 10 cross validation repetitions... 
starting worker pid=8908 on localhost:11120 at 11:10:08.608
starting worker pid=8917 on localhost:11120 at 11:10:08.744
starting worker pid=8926 on localhost:11120 at 11:10:08.881
starting worker pid=8935 on localhost:11120 at 11:10:09.017
starting worker pid=8944 on localhost:11120 at 11:10:09.155
starting worker pid=8953 on localhost:11120 at 11:10:09.291
starting worker pid=8962 on localhost:11120 at 11:10:09.428
Executing 7 processes via parallel... 
Starting cross validation with a total of 10 repetitions... 
Performing repetition 1 out of 10... 
Performing cross validation iteration 1 out of 5... 
Computing the initial values of beta by standard NMF... 
Error in t.default(curr_beta) : argument is not a matrix
In addition: Warning messages:
1: In .Internal(gc(verbose, reset, full)) :
  closing unused connection 9 (<-localhost:11120)
2: In .Internal(gc(verbose, reset, full)) :
  closing unused connection 8 (<-localhost:11120)
3: In .Internal(gc(verbose, reset, full)) :
  closing unused connection 7 (<-localhost:11120)
4: In .Internal(gc(verbose, reset, full)) :
  closing unused connection 6 (<-localhost:11120)
5: In .Internal(gc(verbose, reset, full)) :
  closing unused connection 5 (<-localhost:11120)
6: In .Internal(gc(verbose, reset, full)) :
  closing unused connection 4 (<-localhost:11120)
7: In .Internal(gc(verbose, reset, full)) :
  closing unused connection 3 (<-localhost:11120)

I'm checking the changelog of NMF to understand if some change was introduced in the output of the function basis which seems to be the responsible of this.

[luca-dex]

We're not the first to encounter this error... PoisonAlien/maftools#8

[luca-dex]

The output of this:

curr_beta <- basis(nmf(t(x),rank=K,nrun=nmf_runs))

is NULL

[luca-dex]

Ok, I rewrited the code in this way:

temp_nmf = nmf(t(x),rank=K,nrun=nmf_runs)
curr_beta = basis(temp_nmf)

Now, nmf() is working but the result of this call is not an NMF. As reported in the vignette, when rank is a vector, the result is an object of class NMF.rank that contain all the different object NMF in a list. It is possible to apply the basis function to every object contained in the list but not to the result of the nmf call.

For example, with K = 3:5 I can run this:

temp_nmf = nmf(t(x),rank=K,nrun=nmf_runs)
basis(temp_nmf$fit[['4']])
                           [,1]         [,2]         [,3]         [,4]
M12759_at          2.218067e+03 3.542256e+03 2.872655e-10 7.067707e+04
U46006_s_at        1.461210e+04 5.898066e+02 1.493301e+02 3.914084e+02
X70083_at          6.065541e+02 6.787854e+02 2.159494e+02 5.558119e+02
X03100_cds2_at     1.483671e+05 2.036076e+04 5.522401e+04 5.215328e+04
L32976_at          9.960974e+03 1.547032e+03 3.401427e+03 4.020798e+03
...

[luca-dex]

To solve the other issue with the NMF package, the only way I found is to move it to Depends instead of Imports

[ranasats]

Another partial-solution for this is to load the NMF package along with everything else.

library("SparseSignatures")
library('NMF')
library("BSgenome.Hsapiens.1000genomes.hs37d5")

#3 was my report on this a little bit ago.