Good sklearn

sklearn/inputs.sh

@@ -3,5 +3,6 @@
 cd "$(realpath $(dirname "$0"))"
 mkdir -p tmp
 mkdir -p result
+mkdir -p inputs
 
-# Currently just dumped the entire dataset, but ideally we actually download it
+/usr/bin/env python3 -c "from sklearn.datasets import fetch_kddcup99; fetch_kddcup99(data_home=\"inputs\", percent10=False, download_if_missing=True)"

sklearn/run.sh

@@ -1,8 +1,11 @@
 #!/bin/bash
 
 PYTHON="python3"
-DIR=$PWD
-SCRIPTS=$DIR/scripts
+OUT=${OUT:-$PWD/result}
+TMP=${TMP:-$PWD/tmp}
+#export tmp to env
+export TMP
+SCRIPTS=${SCRIPTS:-$PWD/scripts}
 
 # Ideally, we'll move on to piping rather than writing to a file
 MODEL=$TMP/model.obj

sklearn/scripts/adjust_coef.py

@@ -32,5 +32,6 @@
     else:
         model.intercept_ = np.zeros(n_classes)
 
-with open(destination, 'wb') as file:
+filepath = os.path.join(tmp, 'trained_model.obj')
+with open(filepath, 'wb') as file:
     pickle.dump(model, file)

sklearn/setup.sh

@@ -2,7 +2,7 @@
 
 export PASH_SPEC_TOP=${PASH_SPEC_TOP:-$(git rev-parse --show-toplevel --show-superproject-working-tree)}
 
-benchmark_dir="$PASH_SPEC_TOP/report/benchmarks/sklearn"
+benchmark_dir="sklearn"
 
 cd "$(realpath $(dirname "$0"))"
 mkdir -p "$PASH_SPEC_TOP/report/resources/sklearn"

sklearn/verify.py

@@ -20,7 +20,7 @@
 control_model.fit(X_train, y_train)
 control_score = control_model.score(X_test, y_test)
 
-with open('result/trained_model.obj', 'rb') as file:
+with open('tmp/trained_model.obj', 'rb') as file:
     experiment_model = pickle.load(file)
 experiment_score = experiment_model.score(X_test, y_test)