From 0c07241104119acc29f12f0dd2b8b92b16724d13 Mon Sep 17 00:00:00 2001 From: reedts Date: Mon, 13 Aug 2018 10:02:50 +0200 Subject: [PATCH] Removed unnecessary lines --- pywatts/kcross.py | 16 ---------------- 1 file changed, 16 deletions(-) diff --git a/pywatts/kcross.py b/pywatts/kcross.py index 1b8d54d..9c3151d 100644 --- a/pywatts/kcross.py +++ b/pywatts/kcross.py @@ -35,28 +35,12 @@ def split(data, k): train_samples.extend(samples[j*bucketsize:(j+1)*bucketsize]) # Create new dictionaries in the eval lists - #X_eval.append({'dc': eval_samples[:-1]}) - #y_eval.append({'dc': eval_samples[-1]}) X_eval.append({'dc': [x for s in eval_samples for c, x in enumerate(s, 1) if c % 337 != 0]}) y_eval.append({'dc': [x for s in eval_samples for c, x in enumerate(s, 1) if c % 337 == 0]}) - #X_train.append({'dc': train_samples[:-1]}) - #y_train.append({'dc': train_samples[-1]}) X_train.append({'dc': [x for s in train_samples for c, x in enumerate(s, 1) if c % 337 != 0]}) y_train.append({'dc': [x for s in train_samples for c, x in enumerate(s, 1) if c % 337 == 0]}) - #print(len(X_eval)) - #print(len(y_eval)) - #print(len(X_train)) - #print(len(y_train)) - #print(len(X_train[0]['dc'])) - #print(len(y_train[0]['dc'])) - #print(len(X_eval[0]['dc'])) - #print(len(y_eval[0]['dc'])) - #print(X_train) - #print(y_train) - #exit(0) - return X_train, y_train, X_eval, y_eval