diff --git a/src/fzf.py b/src/fzf.py
index 120e2c7..86c2c11 100644
--- a/src/fzf.py
+++ b/src/fzf.py
@@ -1,126 +1,31 @@
-from __future__ import print_function
-
 import errno
-import os.path
 import subprocess
 import sys
 
-from pkg_resources import resource_exists, resource_filename
-
-__all__ = 'BUNDLED_EXECUTABLE', 'iterfzf'
 
 EXECUTABLE_NAME = 'fzf.exe' if sys.platform == 'win32' else 'fzf'
-BUNDLED_EXECUTABLE = (
-    resource_filename(__name__, EXECUTABLE_NAME)
-    if resource_exists(__name__, EXECUTABLE_NAME)
-    else (
-        os.path.join(os.path.dirname(__file__), EXECUTABLE_NAME)
-        if os.path.isfile(
-            os.path.join(os.path.dirname(__file__), EXECUTABLE_NAME)
-        )
-        else None
-    )
-)
 
 
-def iterfzf(
-    # CHECK: When the signature changes, __init__.pyi file should also change.
-    iterable,
-    # Search mode:
-    extended=True, exact=False, case_sensitive=None,
-    # Interface:
-    multi=False, mouse=True, print_query=False,
-    # Layout:
-    prompt='> ',
-    ansi=None,
-    preview=None,
-    # Misc:
-    query='', encoding=None, executable=BUNDLED_EXECUTABLE or EXECUTABLE_NAME
-):
-    cmd = [executable, '--no-sort', '--prompt=' + prompt]
-    cmd = [executable, '--prompt=' + prompt]
-    if not extended:
-        cmd.append('--no-extended')
-    if case_sensitive is not None:
-        cmd.append('+i' if case_sensitive else '-i')
-    if exact:
-        cmd.append('--exact')
-    if multi:
-        cmd.append('--multi')
-    if not mouse:
-        cmd.append('--no-mouse')
-    if print_query:
-        cmd.append('--print-query')
-    if query:
-        cmd.append('--query=' + query)
-    if preview:
-        cmd.append('--preview=' + preview)
-    if ansi:
-        cmd.append('--ansi')
-    encoding = encoding or sys.getdefaultencoding()
-    proc = None
-    stdin = None
-    byte = None
-    lf = u'\n'
-    cr = u'\r'
-    for line in iterable:
-        if byte is None:
-            byte = isinstance(line, bytes)
-            if byte:
-                lf = b'\n'
-                cr = b'\r'
-        elif isinstance(line, bytes) is not byte:
-            raise ValueError(
-                'element values must be all byte strings or all '
-                'unicode strings, not mixed of them: ' + repr(line)
-            )
-        if lf in line or cr in line:
-            raise ValueError(r"element values must not contain CR({1!r})/"
-                             r"LF({2!r}): {0!r}".format(line, cr, lf))
-        if proc is None:
-            proc = subprocess.Popen(
-                cmd,
-                stdin=subprocess.PIPE,
-                stdout=subprocess.PIPE,
-                stderr=None
-            )
-            stdin = proc.stdin
-        if not byte:
-            line = line.encode(encoding)
-        try:
-            stdin.write(line + b'\n')
-            stdin.flush()
-        except IOError as e:
-            if e.errno != errno.EPIPE and errno.EPIPE != 32:
-                raise
-            break
-    stdin.close()
-    if proc is None or proc.wait() not in [0, 1]:
-        if print_query:
-            return None, None
-        else:
-            return None
+def iterfzf(iterable, prompt='> '):
+    cmd = [EXECUTABLE_NAME, '--prompt=' + prompt]
+    encoding = sys.getdefaultencoding()
+    proc = subprocess.Popen(cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=None)
+    if proc.stdin is None:
+        return None
     try:
-        stdin.close()
+        lines = "\n".join(iterable)
+        proc.stdin.write(lines.encode('utf-8'))
+        proc.stdin.close()
     except IOError as e:
         if e.errno != errno.EPIPE and errno.EPIPE != 32:
             raise
-    stdout = proc.stdout
-    decode = (lambda b: b) if byte else (lambda t: t.decode(encoding))
-    output = [decode(ln.strip(b'\r\n\0')) for ln in iter(stdout.readline, b'')]
-    if print_query:
-        try:
-            if multi:
-                return output[0], output[1:]
-            else:
-                return output[0], output[1]
-        except IndexError:
-            return output[0], None
-    else:
-        if multi:
-            return output
-        else:
-            try:
-                return output[0]
-            except IndexError:
-                return None
+    if proc is None or proc.wait() not in [0, 1]:
+        return None
+    if proc.stdout is None:
+        return None
+    decode = lambda t: t.decode(encoding)
+    output = [decode(ln.strip(b'\r\n\0')) for ln in iter(proc.stdout.readline, b'')]
+    try:
+        return output[0]
+    except IndexError:
+        return None
diff --git a/src/predict.py b/src/predict.py
index 69e5ad3..8482def 100644
--- a/src/predict.py
+++ b/src/predict.py
@@ -5,14 +5,36 @@ from typing import List
 
 def get_categories(transactions: List[Transaction]) -> List[str]:
     categories = set([t.account2 for t in transactions])
-    categories.discard(UNKNOWN_CATEGORY)
+    categories.add(UNKNOWN_CATEGORY)
     return list(categories)
 
 
+def get_sort_categories():
+    def sort_categories(row: str, categories: List[str]):
+        if learn is None:
+            return
+        _, _, probs = learn.predict(row)
+        cat_to_prob = dict(zip(learn.dls.vocab[1],probs.tolist()))
+        categories.sort(key=lambda c: cat_to_prob[c] if c in cat_to_prob else 0.0)
+
+    learn = None
+    try:
+        from fastai.text.all import load_learner
+        learn = load_learner("ldg.pkl")
+    except ModuleNotFoundError:
+        pass
+    except FileNotFoundError:
+        pass
+
+    return sort_categories
+
+
 def add_account2(transactions: List[Transaction]):
     categories = get_categories(transactions)
     unmapped_transactions = filter(lambda t: t.account2 == UNKNOWN_CATEGORY, transactions)
+    sort_categories = get_sort_categories()
     for t in unmapped_transactions:
+        sort_categories(t.row, categories)
         add_account2_interactive(t, categories)