From f65e36925d5c797faa5a48bf1d3820a376c77b26 Mon Sep 17 00:00:00 2001
From: Ines Montani <ines@ines.io>
Date: Tue, 20 Aug 2019 15:08:59 +0200
Subject: [PATCH] Fix absolute imports and avoid importing from cli

---
 spacy/_ml.py             | 15 +++++++++++++++
 spacy/cli/pretrain.py    | 19 ++-----------------
 spacy/pipeline/pipes.pyx |  6 ++----
 3 files changed, 19 insertions(+), 21 deletions(-)

diff --git a/spacy/_ml.py b/spacy/_ml.py
index 0411c4bd4..660d20c46 100644
--- a/spacy/_ml.py
+++ b/spacy/_ml.py
@@ -808,3 +808,18 @@ def _replace_word(word, random_words, mask="[MASK]"):
         return random_words.next()
     else:
         return word
+
+
+def get_cossim_loss(yh, y):
+    # Add a small constant to avoid 0 vectors
+    yh = yh + 1e-8
+    y = y + 1e-8
+    # https://math.stackexchange.com/questions/1923613/partial-derivative-of-cosine-similarity
+    xp = get_array_module(yh)
+    norm_yh = xp.linalg.norm(yh, axis=1, keepdims=True)
+    norm_y = xp.linalg.norm(y, axis=1, keepdims=True)
+    mul_norms = norm_yh * norm_y
+    cosine = (yh * y).sum(axis=1, keepdims=True) / mul_norms
+    d_yh = (y / mul_norms) - (cosine * (yh / norm_yh ** 2))
+    loss = xp.abs(cosine - 1).sum()
+    return loss, -d_yh
diff --git a/spacy/cli/pretrain.py b/spacy/cli/pretrain.py
index 99ea5dadc..e78cdc9b8 100644
--- a/spacy/cli/pretrain.py
+++ b/spacy/cli/pretrain.py
@@ -10,7 +10,7 @@ from collections import Counter
 from pathlib import Path
 from thinc.v2v import Affine, Maxout
 from thinc.misc import LayerNorm as LN
-from thinc.neural.util import prefer_gpu, get_array_module
+from thinc.neural.util import prefer_gpu
 from wasabi import Printer
 import srsly
 
@@ -18,7 +18,7 @@ from ..errors import Errors
 from ..tokens import Doc
 from ..attrs import ID, HEAD
 from .._ml import Tok2Vec, flatten, chain, create_default_optimizer
-from .._ml import masked_language_model
+from .._ml import masked_language_model, get_cossim_loss
 from .. import util
 from .train import _load_pretrained_tok2vec
 
@@ -307,21 +307,6 @@ def get_vectors_loss(ops, docs, prediction, objective="L2"):
     return loss, d_target
 
 
-def get_cossim_loss(yh, y):
-    # Add a small constant to avoid 0 vectors
-    yh = yh + 1e-8
-    y = y + 1e-8
-    # https://math.stackexchange.com/questions/1923613/partial-derivative-of-cosine-similarity
-    xp = get_array_module(yh)
-    norm_yh = xp.linalg.norm(yh, axis=1, keepdims=True)
-    norm_y = xp.linalg.norm(y, axis=1, keepdims=True)
-    mul_norms = norm_yh * norm_y
-    cosine = (yh * y).sum(axis=1, keepdims=True) / mul_norms
-    d_yh = (y / mul_norms) - (cosine * (yh / norm_yh ** 2))
-    loss = xp.abs(cosine - 1).sum()
-    return loss, -d_yh
-
-
 def create_pretraining_model(nlp, tok2vec):
     """Define a network for the pretraining. We simply add an output layer onto
     the tok2vec input model. The tok2vec input model needs to be a model that
diff --git a/spacy/pipeline/pipes.pyx b/spacy/pipeline/pipes.pyx
index db45e0faa..8b15ebd38 100644
--- a/spacy/pipeline/pipes.pyx
+++ b/spacy/pipeline/pipes.pyx
@@ -13,9 +13,6 @@ from thinc.misc import LayerNorm
 from thinc.neural.util import to_categorical
 from thinc.neural.util import get_array_module
 
-from spacy.kb import KnowledgeBase
-
-from spacy.cli.pretrain import get_cossim_loss
 from .functions import merge_subtokens
 from ..tokens.doc cimport Doc
 from ..syntax.nn_parser cimport Parser
@@ -27,7 +24,8 @@ from ..vocab cimport Vocab
 from ..syntax import nonproj
 from ..attrs import POS, ID
 from ..parts_of_speech import X
-from .._ml import Tok2Vec, build_tagger_model, cosine
+from ..kb import KnowledgeBase
+from .._ml import Tok2Vec, build_tagger_model, cosine, get_cossim_loss
 from .._ml import build_text_classifier, build_simple_cnn_text_classifier
 from .._ml import build_bow_text_classifier, build_nel_encoder
 from .._ml import link_vectors_to_models, zero_init, flatten