prep for release

CHANGES.md

@@ -1,9 +1,10 @@
-v0.2.7 (dev)
+v0.2.7
 ======
 + new subcommand `ancestry` to predict ancestry using a simple neural network on the somalier
   sketches. creates an interactive html output and a text file
 + fix for "Argument list too long" on huge cohorts (#37)
 + sub-sample .pairs.tsv output for huge cohorts -- only for unrelated samples.
++ better sub-sampling of html output
 
 v0.2.6
 ======

README.md

@@ -30,6 +30,9 @@ to add/remove samples or adjust a pedigree file and re-run iteratively.
 For example to add the **n + 1th** samples, just run `somalier extract` on the new sample and then re-use
 the already extracted data from the `n` original samples.
 
+For *huge* sample-sets, if you run into a bash error for *argument list too long*, you can pass the somalier files as quoted
+glob strings like:  `"/path/to/set-a/*.somalier" "/path/to/set-b/*.somalier"`.
+
 ## Usage
 
 The usage is also described above. Briefly, run:

src/somalier.nim

@@ -232,7 +232,7 @@ proc main() =
     "extract": pair(f:extract_main, description: "extract genotype-like information for a single sample from VCF/BAM/CRAM."),
     "relate": pair(f:rel_main, description: "aggregate `extract`ed information and calculate relatedness among samples."),
     "ancestry": pair(f:ancestry_main, description: "perform ancestry prediction on a set of samples, given a set of labeled samples"),
-    "depthview": pair(f:depth_main, description: "plot per-chromosome depth for each sample for quick quality-control"),
+    #"depthview": pair(f:depth_main, description: "plot per-chromosome depth for each sample for quick quality-control"),
     "find-sites": pair(f:findsites_main, description: "create a new sites.vcf.gz file from a population VCF (this is rarely needed)."),
   }.toOrderedTable
 

src/somalierpkg/ancestry.nim

@@ -52,6 +52,11 @@ type ForHtml = ref object
   probs: seq[float32] # probability of maximum prediction
   ancestry_label: string
 
+proc subset(T: var Tensor[float32], Q: var Tensor[float32], labels: var Tensor[int]) =
+  echo T.shape
+  echo Q.shape
+  echo labels.shape
+
 proc ancestry_main*() =
 
   var argv = commandLineParams()
@@ -117,13 +122,17 @@ proc ancestry_main*() =
       vec[j] = ac.ab(5).alts.float32
     query_mat[i] = vec
 
+
   var
     nPCs = parseInt(opts.n_pcs)
     T = train_mat.toTensor()
+    Q = query_mat.toTensor()
     Y = int_labels.toTensor() #.astype(float32)#.unsqueeze(0).transpose
     t0 = cpuTime()
     res = T.pca(nPCs) #, center=true) #, n_power_iters=4)
 
+  #subset(T, Q, Y)
+
   stderr.write_line &"[somalier] time for dimensionality reduction to shape {res.projected.shape}: {cpuTime() - t0:.2f} seconds"
 
   let
@@ -187,7 +196,6 @@ proc ancestry_main*() =
     let t_probs = model.forward(X).value.softmax #.argmax(axis=1).squeeze
 
   let
-    Q = query_mat.toTensor()
     q_proj = Q * res.components
     q_probs = model.forward(ctx.variable q_proj).value.softmax
     q_pred = q_probs.argmax(axis=1).squeeze

src/somalierpkg/results.html

@@ -157,18 +157,18 @@ <h5>Sample Depth Metrics</h5>
 var sample_data = <SAMPLE_JSON>
 var input = <INPUT_JSON>
 
-var colors = ['rgba(55,126,184,0.7)', 'rgba(228,26,28,0.7)', 'rgba(77,175,74,0.7)', 'rgba(152,78,163,0.7)', 'rgba(255,127,0,0.7)', 'rgba(166,86,40,0.7)', 'rgba(247,129,191,0.7)']
+var colors = ['rgba(55,126,184,0.7)', 'rgba(228,26,28,0.7)', 'rgba(152,78,163,0.7)', 'rgba(255,127,0,0.7)', 'rgba(166,86,40,0.7)', 'rgba(247,129,191,0.7)', 'rgba(77,175,74,0.7)',]
 var size
 if (sample_data.length > 700) {
-    size = 6
+    size = 8
 } else if (sample_data.length > 200) {
-    size = 7
+    size = 9
 } else if (sample_data.length > 50) {
-    size = 8
-} else if (sample_data.length > 20) {
     size = 10
-} else {
+} else if (sample_data.length > 20) {
     size = 12
+} else {
+    size = 15
 }
 
 function set_xy_data_by_group(input, metric, is_x) {