Hf integration

conf/test_noise_eval.config

@@ -31,6 +31,10 @@ params {
     save_data    = false
 }
 
+env {
+    HF_DATASETS_CACHE = '/tmp/hf_cache'
+}
+
 // Limit resources so that this can run on GitHub Actions
 process {
     maxRetries = params.max_retries

modules/local/stimulus/check_model/main.nf

@@ -19,7 +19,6 @@ process CHECK_MODEL {
     def args = task.ext.args ?: ''
     """
     stimulus check-model \
-        -e ${data_config} \
         -d ${data} \
         -m ${model} \
         -c ${model_config} \

modules/local/stimulus/compare_tensors/main.nf

@@ -1,5 +1,5 @@
 process STIMULUS_COMPARE_TENSORS {
-    tag "${meta.id}"
+    tag "${meta.id1}"
     label 'process_medium'
     container "docker.io/mathysgrapotte/stimulus-py:dev"
 
@@ -18,7 +18,7 @@ process STIMULUS_COMPARE_TENSORS {
     """
     stimulus compare-tensors \
         ${tensors} \
-        -s scores.csv \
+        -o scores.csv \
         ${args}
 
     # Extract first row of scores.csv

modules/local/stimulus/encode/main.nf

@@ -0,0 +1,41 @@
+process ENCODE_CSV {
+
+    tag "${meta.id}"
+    label 'process_medium'
+    // TODO: push image to nf-core quay.io
+    container "docker.io/mathysgrapotte/stimulus-py:dev"
+
+    input:
+    tuple val(meta), path(data)
+    tuple val(meta2), path(config)
+
+    output:
+    tuple val(meta2), path("${prefix}_encoded"), emit: encoded
+    path "versions.yml"          , emit: versions
+
+    script:
+    def args = task.ext.args ?: ''
+    prefix = task.ext.prefix ?: "${meta.split_id}-${meta2.transform_id}"
+    """
+    stimulus encode-csv \
+        -d ${data} \
+        -y ${config} \
+        -o ${prefix}_encoded \
+        $args
+
+    cat <<-END_VERSIONS > versions.yml
+    "${task.process}":
+        stimulus: \$(stimulus -v | cut -d ' ' -f 3)
+    END_VERSIONS
+    """
+
+    stub:
+    """
+    echo passing check-model stub
+
+    cat <<-END_VERSIONS > versions.yml
+    "${task.process}":
+        stimulus: \$(stimulus -v | cut -d ' ' -f 3)
+    END_VERSIONS
+    """
+}

modules/local/stimulus/predict/main.nf

@@ -5,7 +5,7 @@ process STIMULUS_PREDICT {
 
     input:
     tuple val(meta) , path(model), path(model_config), path(weigths)
-    tuple val(meta2), path(data), path(data_config)
+    tuple val(meta2), path(data), path(config)
 
     output:
     tuple val(meta), path("${prefix}-pred.safetensors"), emit: predictions
@@ -17,7 +17,6 @@ process STIMULUS_PREDICT {
     """
     stimulus predict \
         -d ${data} \
-        -e ${data_config} \
         -m ${model} \
         -c ${model_config} \
         -w ${weigths} \

modules/local/stimulus/split_csv/main.nf

@@ -10,7 +10,7 @@ process STIMULUS_SPLIT_DATA {
     tuple val(meta2), path(sub_config)
 
     output:
-    tuple val(meta2), path("${prefix}.csv"), emit: csv_with_split
+    tuple val(meta2), path("${prefix}_split"), emit: csv_with_split
     path "versions.yml"          , emit: versions
 
     script:
@@ -19,7 +19,7 @@ process STIMULUS_SPLIT_DATA {
     stimulus split-csv \
         -c ${data} \
         -y ${sub_config} \
-        -o ${prefix}.csv
+        -o ${prefix}_split
 
     cat <<-END_VERSIONS > versions.yml
     "${task.process}":
@@ -30,7 +30,7 @@ process STIMULUS_SPLIT_DATA {
     stub:
     prefix = task.ext.prefix ?: "${meta.id}-split-${meta2.id}"
     """
-    touch ${prefix}.csv
+    touch ${prefix}_split
 
     cat <<-END_VERSIONS > versions.yml
     "${task.process}":

modules/local/stimulus/transform_csv/main.nf

@@ -11,7 +11,7 @@ process STIMULUS_TRANSFORM_CSV {
     tuple val(meta2), path(config)
 
     output:
-    tuple val(meta), path("${prefix}.csv"), emit: transformed_data
+    tuple val(meta), path("${prefix}"), emit: transformed_data
     path "versions.yml"          , emit: versions
 
     script:
@@ -20,7 +20,7 @@ process STIMULUS_TRANSFORM_CSV {
     stimulus transform-csv \
         -c ${data} \
         -y ${config} \
-        -o ${prefix}.csv
+        -o ${prefix}
 
     cat <<-END_VERSIONS > versions.yml
     "${task.process}":

modules/local/stimulus/tune/main.nf

@@ -25,7 +25,6 @@ process STIMULUS_TUNE {
     stimulus tune \
         -d ${transformed_data} \
         -m ${model} \
-        -e ${data_sub_config} \
         -c ${model_config} \
         -o ${prefix}-best-model.safetensors \
         -bo ${prefix}-best-optimizer.opt \

subworkflows/local/evaluation/main.nf

@@ -28,10 +28,9 @@ workflow EVALUATION_WF {
     // Evaluation mode 1: Predict the data using the best model
     // and then compare the predictions of 2 different models
     //
-
     STIMULUS_PREDICT(
         model,
-        ch_data.collect()
+        ch_data.first() // converts a queue channel to a value channel
     )
     ch_versions = ch_versions.mix(STIMULUS_PREDICT.out.versions)
     predictions = STIMULUS_PREDICT.out.predictions
@@ -41,26 +40,42 @@ workflow EVALUATION_WF {
     // and the same number of trials, we can estimate the noise across replicates
     // This is done by comparing the predictions of the alternative models between each other
     // and then calculatin a summary metric over them (e.g. mean, median, std, etc.)
-
-    replicate_predictions = predictions.map{
-                meta, prediction ->
-                    [["id": meta.id,
-                    "split_id": meta.split_id,
-                    "transform_id": meta.transform_id,
-                    "n_trials": meta.n_trials ], meta, prediction]
-            }.groupTuple(by:0)
-            .map{
-                merging_meta, metas, predictions ->
-                    [merging_meta, predictions]
+    pairs = predictions
+        .collate(2)
+        .collect()
+        .map { items ->
+            def pairs = []
+            // Create all unique combinations using index comparison
+            (0..<items.size()).each { i ->
+                (i+1..<items.size()).each { j ->
+                    def meta1 = items[i][0]
+                    def meta2 = items[j][0]
+                    def files = [items[i][1], items[j][1]]
+                    // Only compare different transforms OR different replicates
+                    if(meta1.transform_id != meta2.transform_id || meta1.replicate != meta2.replicate) {
+                        pairs << [
+                            [
+                                "id1": meta1.id,
+                                "id2": meta2.id,
+                                "split_id1": meta1.split_id,
+                                "split_id2": meta2.split_id,
+                                "transform_id1": meta1.transform_id,
+                                "transform_id2": meta2.transform_id,
+                                "replicate1": meta1.replicate,
+                                "replicate2": meta2.replicate
+                            ],
+                            // Create unique filenames using both transforms and replicates
+                            files
+                        ]
+                    }
+                }
             }
-
-    // check if the predictions are at least 2, meta,predictions
-    replicate_predictions.filter{
-        it[1].size() > 1
-    }.set{ replicate_predictions }
+            pairs
+        }
+        .flatMap { it }
 
     STIMULUS_COMPARE_TENSORS_COSINE(
-        replicate_predictions
+        pairs
     )
 
     cosine_scores = STIMULUS_COMPARE_TENSORS_COSINE.out.csv

subworkflows/local/transform_csv/main.nf

@@ -5,6 +5,7 @@
 */
 
 include { STIMULUS_TRANSFORM_CSV } from '../../../modules/local/stimulus/transform_csv'
+include { ENCODE_CSV } from '../../../modules/local/stimulus/encode'
 
 /*
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -48,17 +49,23 @@ workflow TRANSFORM_CSV_WF {
             config:
             [meta, config]
         }
-
     // run stimulus transform
     STIMULUS_TRANSFORM_CSV(
         ch_input.data,
         ch_input.config
     )
     ch_transformed_data = STIMULUS_TRANSFORM_CSV.out.transformed_data
-    ch_versions = ch_versions.mix(STIMULUS_TRANSFORM_CSV.out.versions)
+
+    // run stimulus encode
+    ENCODE_CSV(
+        ch_transformed_data,
+        ch_input.config
+    )
+    ch_encoded_data = ENCODE_CSV.out.encoded
+    ch_versions = ch_versions.mix(ENCODE_CSV.out.versions)
 
     emit:
-    transformed_data = ch_transformed_data
+    transformed_data = ch_encoded_data
     versions = ch_versions // channel: [ versions.yml ]
 }
 

workflows/deepmodeloptim.nf

@@ -16,6 +16,7 @@ include { SPLIT_CSV_WF                        } from '../subworkflows/local/spli
 include { TRANSFORM_CSV_WF                    } from '../subworkflows/local/transform_csv'
 include { TUNE_WF                             } from '../subworkflows/local/tune'
 include { EVALUATION_WF                       } from '../subworkflows/local/evaluation'
+include { ENCODE_CSV                          } from '../modules/local/stimulus/encode'
 
 //
 // MODULES: Consisting of nf-core/modules
@@ -149,6 +150,11 @@ workflow DEEPMODELOPTIM {
     // Now the data config will not work if passed in full
     // We need to pass in the split data config, any of them, for the predict modules
     // This will be changed in the future
+    ENCODE_CSV(
+        prediction_data,
+        TUNE_WF.out.data_config_tmp.first()
+    )
+    prediction_data = ENCODE_CSV.out.encoded
     prediction_data = prediction_data.combine(TUNE_WF.out.data_config_tmp.first().map{meta,file -> file})
     EVALUATION_WF(
         TUNE_WF.out.model_tmp,