Massive speed increase. Added latent caching both to disk and to memory

toolkit/basic.py

@@ -1,4 +1,13 @@
+import gc
+
+import torch
 
 
 def value_map(inputs, min_in, max_in, min_out, max_out):
     return (inputs - min_in) * (max_out - min_out) / (max_in - min_in) + min_out
+
+
+def flush(garbage_collect=True):
+    torch.cuda.empty_cache()
+    if garbage_collect:
+        gc.collect()