diff --git a/README.md b/README.md
index e52dcdb..a75001a 100644
--- a/README.md
+++ b/README.md
@@ -11,5 +11,9 @@ Additionally, `pip install -r optrequirements.txt` for optional dependencies use
 
 ## Building
 
-`python run_mindep.py run lang1 lang2 ... langn` for an initial run.
+```
+python run_mindep.py run lang1 lang2 ... langn > out.csv
+python run_mindep.py postprocess out.csv > processed.csv
+```
+
 The list of langs can be found at `corpora.ud_langs`.
diff --git a/mindep/__init__.py b/mindep/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/mindep.py b/mindep/mindep.py
similarity index 100%
rename from mindep.py
rename to mindep/mindep.py
diff --git a/opt_mindep.py b/mindep/opt_mindep.py
similarity index 99%
rename from opt_mindep.py
rename to mindep/opt_mindep.py
index 9b94192..3a65cbe 100644
--- a/opt_mindep.py
+++ b/mindep/opt_mindep.py
@@ -9,7 +9,7 @@
 from rfutils import mreduce_by_key
 
 import conditioning as cond
-import mindep
+from . import mindep
 
 WEIGHT_MIN = -1
 WEIGHT_MAX = 1
diff --git a/run_mindep.py b/run_mindep.py
index 8c69625..9d43f97 100644
--- a/run_mindep.py
+++ b/run_mindep.py
@@ -1,7 +1,6 @@
 from __future__ import division
 import sys
 import copy
-import itertools
 import functools
 import random
 import csv
@@ -10,8 +9,8 @@
 import pandas as pd
 #from distributed import Executor, as_completed
 
-import mindep
-import opt_mindep
+from mindep import mindep
+from mindep import opt_mindep
 #import linearize as lin
 import corpora
 
@@ -32,6 +31,7 @@ def with_open(filename, mode, f):
 
 @rfutils.memoize
 def load_linearization_model(lang, spec):
+    import pickle
     return with_open(MODEL_FILENAME_TEMPLATE % (lang, spec), 'rb', pickle.load)
 
 LANGS = set(corpora.ud_corpora.keys())