def test_predict_tags_mesh_cnn(mesh_cnn_path, mesh_label_binarizer_path):
tags = predict_tags(X,
mesh_cnn_path,
mesh_label_binarizer_path,
approach="mesh-cnn")
assert len(tags) == 5
tags = predict_tags(X,
mesh_cnn_path,
mesh_label_binarizer_path,
approach="mesh-cnn",
probabilities=True)
for tags_ in tags:
for tag, prob in tags_.items():
assert 0 <= prob <= 1.0
tags = predict_tags(X,
mesh_cnn_path,
mesh_label_binarizer_path,
approach="mesh-cnn",
threshold=0)
for tags_ in tags:
assert len(tags_) == 5000
tags = predict_tags(X,
mesh_cnn_path,
mesh_label_binarizer_path,
approach="mesh-cnn",
threshold=1)
for tags_ in tags:
assert len(tags_) == 0
def test_predict_tags_science_ensemble(science_ensemble_path,
label_binarizer_path):
tags = predict_tags(X,
model_path=science_ensemble_path,
label_binarizer_path=label_binarizer_path,
approach="science-ensemble")
assert len(tags) == 5
tags = predict_tags(X,
model_path=science_ensemble_path,
label_binarizer_path=label_binarizer_path,
approach="science-ensemble",
probabilities=True)
for tags_ in tags:
for tag, prob in tags_.items():
assert 0 <= prob <= 1.0
tags = predict_tags(X,
model_path=science_ensemble_path,
label_binarizer_path=label_binarizer_path,
approach="science-ensemble",
threshold=0)
for tags_ in tags:
assert len(tags_) == 24
tags = predict_tags(X,
model_path=science_ensemble_path,
label_binarizer_path=label_binarizer_path,
approach="science-ensemble",
threshold=1)
for tags_ in tags:
assert len(tags_) == 0
def test_predict_tags_tfidf_svm(tfidf_svm_path, label_binarizer_path):
tags = predict_tags(X,
model_path=tfidf_svm_path,
label_binarizer_path=label_binarizer_path,
approach="tfidf-svm")
assert len(tags) == 5
tags = predict_tags(X,
model_path=tfidf_svm_path,
label_binarizer_path=label_binarizer_path,
approach="tfidf-svm",
probabilities=True)
for tags_ in tags:
for tag, prob in tags_.items():
assert 0 <= prob <= 1.0
tags = predict_tags(X,
model_path=tfidf_svm_path,
label_binarizer_path=label_binarizer_path,
approach="tfidf-svm",
threshold=0)
for tags_ in tags:
assert len(tags_) == 24
tags = predict_tags(X,
model_path=tfidf_svm_path,
label_binarizer_path=label_binarizer_path,
approach="tfidf-svm",
threshold=1)
for tags_ in tags:
assert len(tags_) == 0
def test_predict_tags_mesh_xlinear(mesh_xlinear_path,
mesh_label_binarizer_path):
# We need to pass parameters because the load function is different
# depending on the vectorizer library (pecos or sklearn)
parameters = str({'vectorizer_library': 'sklearn'})
tags = predict_tags(X,
mesh_xlinear_path,
mesh_label_binarizer_path,
approach="mesh-xlinear",
parameters=parameters)
assert len(tags) == 5
tags = predict_tags(X,
mesh_xlinear_path,
mesh_label_binarizer_path,
approach="mesh-xlinear",
parameters=parameters,
probabilities=True)
for tags_ in tags:
for tag, prob in tags_.items():
assert 0 <= prob <= 1.0
tags = predict_tags(X,
mesh_xlinear_path,
mesh_label_binarizer_path,
approach="mesh-xlinear",
threshold=0,
parameters=parameters)
for tags_ in tags:
assert len(tags_) == 5000
tags = predict_tags(X,
mesh_xlinear_path,
mesh_label_binarizer_path,
approach="mesh-xlinear",
threshold=1,
parameters=parameters)
for tags_ in tags:
assert len(tags_) == 0
def tag_grants(grants_path,
tagged_grants_path,
model_path,
label_binarizer_path,
approach,
threshold=0.5,
grant_id_field="grant_id",
grant_text_fields=["title", "synopsis"],
text_null_value="No Data Entered"):
with open(tagged_grants_path, "w") as tagged_grants_tf:
fieldnames = ["Grant id", "Tag", "Prob"]
csv_writer = csv.DictWriter(tagged_grants_tf,
fieldnames=fieldnames,
delimiter=",",
quotechar='"',
quoting=csv.QUOTE_MINIMAL)
csv_writer.writeheader()
for grants in yield_batched_grants(grants_path, 128):
grants_text = [
" ".join([
grant[field].replace(text_null_value, "")
for field in grant_text_fields
]) for grant in grants
]
# Removes consecutive white spaces which are uninformative and may cause error #30
grants_text = [
" ".join(text.split()) for text in grants_text if text.strip()
] # Removes empty text
grants_tags = predict_tags(grants_text,
model_path,
label_binarizer_path,
approach,
probabilities=True,
threshold=threshold)
for grant, tags in zip(grants, grants_tags):
for tag, prob in tags.items():
csv_writer.writerow({
'Grant id': grant[grant_id_field],
'Tag': tag,
'Prob': prob
})
tagged_grants_tf.flush()
},
}
model_option = st.sidebar.selectbox("Model", options=list(models.keys()))
model = models[model_option]
probabilities = st.sidebar.checkbox("Display probabilities")
if text == DEFAULT_TEXT:
st.stop()
with st.spinner("Calculating tags..."):
tags = predict_tags(
[text],
model["model_path"],
model["label_binarizer_path"],
model["approach"],
probabilities=probabilities,
threshold=threshold,
)
tags = tags[0]
st.success("Done!")
if probabilities:
tag_probs = [
{"Tag": tag, "Prob": prob} for tag, prob in tags.items() if prob > threshold
]
st.table(pd.DataFrame(tag_probs))
tags = [tag_prob["Tag"] for tag_prob in tag_probs]
else:
for tag in tags:
st.button(tag)
