Spaces:
Runtime error
Runtime error
Merge pull request #9 from ugm2/fix/indexing
Browse files- .streamlit/config.toml +1 -1
- app.py +1 -1
- core/pipelines.py +3 -4
- core/search_index.py +7 -1
- interface/components.py +21 -24
- interface/config.py +5 -1
- interface/pages.py +6 -3
.streamlit/config.toml
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
[theme]
|
2 |
-
primaryColor="#
|
3 |
backgroundColor="#0e1117"
|
4 |
secondaryBackgroundColor="#282929"
|
5 |
textColor = "#ffffff"
|
|
|
1 |
[theme]
|
2 |
+
primaryColor="#e5ab00"
|
3 |
backgroundColor="#0e1117"
|
4 |
secondaryBackgroundColor="#282929"
|
5 |
textColor = "#ffffff"
|
app.py
CHANGED
@@ -29,7 +29,7 @@ def run_demo():
|
|
29 |
with navigation:
|
30 |
|
31 |
selected_page = option_menu(
|
32 |
-
menu_title=
|
33 |
options=list(pages.keys()),
|
34 |
icons=[f[1] for f in pages.values()],
|
35 |
menu_icon="cast",
|
|
|
29 |
with navigation:
|
30 |
|
31 |
selected_page = option_menu(
|
32 |
+
menu_title=None,
|
33 |
options=list(pages.keys()),
|
34 |
icons=[f[1] for f in pages.values()],
|
35 |
menu_icon="cast",
|
core/pipelines.py
CHANGED
@@ -19,6 +19,8 @@ def keyword_search(index="documents", split_word_length=100):
|
|
19 |
|
20 |
- Documents that have more lexical overlap with the query are more likely to be relevant
|
21 |
- Words that occur in fewer documents are more significant than words that occur in many documents
|
|
|
|
|
22 |
"""
|
23 |
document_store = InMemoryDocumentStore(index=index)
|
24 |
keyword_retriever = TfidfRetriever(document_store=(document_store))
|
@@ -39,10 +41,7 @@ def keyword_search(index="documents", split_word_length=100):
|
|
39 |
index_pipeline = Pipeline()
|
40 |
index_pipeline.add_node(processor, name="Preprocessor", inputs=["File"])
|
41 |
index_pipeline.add_node(
|
42 |
-
|
43 |
-
)
|
44 |
-
index_pipeline.add_node(
|
45 |
-
document_store, name="DocumentStore", inputs=["TfidfRetriever"]
|
46 |
)
|
47 |
|
48 |
return search_pipeline, index_pipeline
|
|
|
19 |
|
20 |
- Documents that have more lexical overlap with the query are more likely to be relevant
|
21 |
- Words that occur in fewer documents are more significant than words that occur in many documents
|
22 |
+
|
23 |
+
:warning: **(HAYSTACK BUG) Keyword Search doesn't work if you reindex:** Please refresh page in order to reindex
|
24 |
"""
|
25 |
document_store = InMemoryDocumentStore(index=index)
|
26 |
keyword_retriever = TfidfRetriever(document_store=(document_store))
|
|
|
41 |
index_pipeline = Pipeline()
|
42 |
index_pipeline.add_node(processor, name="Preprocessor", inputs=["File"])
|
43 |
index_pipeline.add_node(
|
44 |
+
document_store, name="DocumentStore", inputs=["Preprocessor"]
|
|
|
|
|
|
|
45 |
)
|
46 |
|
47 |
return search_pipeline, index_pipeline
|
core/search_index.py
CHANGED
@@ -1,4 +1,5 @@
|
|
1 |
from haystack.schema import Document
|
|
|
2 |
import uuid
|
3 |
|
4 |
|
@@ -17,8 +18,12 @@ def format_docs(documents):
|
|
17 |
return db_docs, [doc.meta["id"] for doc in db_docs]
|
18 |
|
19 |
|
20 |
-
def index(documents, pipeline):
|
21 |
documents, doc_ids = format_docs(documents)
|
|
|
|
|
|
|
|
|
22 |
pipeline.run(documents=documents)
|
23 |
return doc_ids
|
24 |
|
@@ -38,6 +43,7 @@ def search(queries, pipeline):
|
|
38 |
"score": res.score,
|
39 |
"id": res.meta["id"],
|
40 |
"fragment_id": res.id,
|
|
|
41 |
}
|
42 |
)
|
43 |
if not score_is_empty:
|
|
|
1 |
from haystack.schema import Document
|
2 |
+
from haystack.document_stores import BaseDocumentStore
|
3 |
import uuid
|
4 |
|
5 |
|
|
|
18 |
return db_docs, [doc.meta["id"] for doc in db_docs]
|
19 |
|
20 |
|
21 |
+
def index(documents, pipeline, clear_index=True):
|
22 |
documents, doc_ids = format_docs(documents)
|
23 |
+
if clear_index:
|
24 |
+
document_stores = pipeline.get_nodes_by_class(class_type=BaseDocumentStore)
|
25 |
+
for docstore in document_stores:
|
26 |
+
docstore.delete_index(docstore.index)
|
27 |
pipeline.run(documents=documents)
|
28 |
return doc_ids
|
29 |
|
|
|
43 |
"score": res.score,
|
44 |
"id": res.meta["id"],
|
45 |
"fragment_id": res.id,
|
46 |
+
"meta": res.meta,
|
47 |
}
|
48 |
)
|
49 |
if not score_is_empty:
|
interface/components.py
CHANGED
@@ -42,11 +42,15 @@ def component_select_pipeline(container):
|
|
42 |
"index_pipeline": index_pipeline,
|
43 |
"doc": pipeline_funcs[index_pipe].__doc__,
|
44 |
}
|
|
|
45 |
|
46 |
|
47 |
def component_show_pipeline(pipeline, pipeline_name):
|
48 |
"""Draw the pipeline"""
|
49 |
-
|
|
|
|
|
|
|
50 |
if pipeline["doc"] is not None:
|
51 |
st.markdown(pipeline["doc"])
|
52 |
fig = get_pipeline_graph(pipeline[pipeline_name])
|
@@ -59,41 +63,39 @@ def component_show_search_result(container, results):
|
|
59 |
st.markdown(f"### Match {idx+1}")
|
60 |
st.markdown(f"**Text**: {document['text']}")
|
61 |
st.markdown(f"**Document**: {document['id']}")
|
|
|
|
|
62 |
if document["score"] is not None:
|
63 |
st.markdown(f"**Score**: {document['score']:.3f}")
|
64 |
st.markdown("---")
|
65 |
|
66 |
|
67 |
-
def component_text_input(container):
|
68 |
"""Draw the Text Input widget"""
|
69 |
with container:
|
70 |
texts = []
|
71 |
-
doc_id = 1
|
72 |
with st.expander("Enter documents"):
|
73 |
while True:
|
74 |
text = st.text_input(f"Document {doc_id}", key=doc_id)
|
75 |
if text != "":
|
76 |
-
texts.append({"text": text})
|
77 |
doc_id += 1
|
78 |
st.markdown("---")
|
79 |
else:
|
80 |
break
|
81 |
-
corpus = [
|
82 |
-
|
83 |
-
]
|
84 |
-
return corpus
|
85 |
|
86 |
|
87 |
-
def component_article_url(container):
|
88 |
"""Draw the Article URL widget"""
|
89 |
with container:
|
90 |
urls = []
|
91 |
-
doc_id = 1
|
92 |
with st.expander("Enter URLs"):
|
93 |
while True:
|
94 |
url = st.text_input(f"URL {doc_id}", key=doc_id)
|
95 |
if url != "":
|
96 |
-
urls.append({"text": extract_text_from_url(url)})
|
97 |
doc_id += 1
|
98 |
st.markdown("---")
|
99 |
else:
|
@@ -101,19 +103,16 @@ def component_article_url(container):
|
|
101 |
|
102 |
for idx, doc in enumerate(urls):
|
103 |
with st.expander(f"Preview URL {idx}"):
|
104 |
-
st.write(doc)
|
105 |
|
106 |
-
corpus = [
|
107 |
-
|
108 |
-
]
|
109 |
-
return corpus
|
110 |
|
111 |
|
112 |
-
def component_file_input(container):
|
113 |
"""Draw the extract text from file widget"""
|
114 |
with container:
|
115 |
files = []
|
116 |
-
doc_id = 1
|
117 |
with st.expander("Enter Files"):
|
118 |
while True:
|
119 |
file = st.file_uploader(
|
@@ -122,7 +121,7 @@ def component_file_input(container):
|
|
122 |
if file != None:
|
123 |
extracted_text = extract_text_from_file(file)
|
124 |
if extracted_text != None:
|
125 |
-
files.append({"text": extracted_text})
|
126 |
doc_id += 1
|
127 |
st.markdown("---")
|
128 |
else:
|
@@ -132,9 +131,7 @@ def component_file_input(container):
|
|
132 |
|
133 |
for idx, doc in enumerate(files):
|
134 |
with st.expander(f"Preview File {idx}"):
|
135 |
-
st.write(doc)
|
136 |
|
137 |
-
corpus = [
|
138 |
-
|
139 |
-
]
|
140 |
-
return corpus
|
|
|
42 |
"index_pipeline": index_pipeline,
|
43 |
"doc": pipeline_funcs[index_pipe].__doc__,
|
44 |
}
|
45 |
+
st.session_state["doc_id"] = 0
|
46 |
|
47 |
|
48 |
def component_show_pipeline(pipeline, pipeline_name):
|
49 |
"""Draw the pipeline"""
|
50 |
+
expander_text = "Show pipeline"
|
51 |
+
if pipeline["doc"] is not None and "BUG" in pipeline["doc"]:
|
52 |
+
expander_text += " ⚠️"
|
53 |
+
with st.expander(expander_text):
|
54 |
if pipeline["doc"] is not None:
|
55 |
st.markdown(pipeline["doc"])
|
56 |
fig = get_pipeline_graph(pipeline[pipeline_name])
|
|
|
63 |
st.markdown(f"### Match {idx+1}")
|
64 |
st.markdown(f"**Text**: {document['text']}")
|
65 |
st.markdown(f"**Document**: {document['id']}")
|
66 |
+
if "_split_id" in document["meta"]:
|
67 |
+
st.markdown(f"**Document Chunk**: {document['meta']['_split_id']}")
|
68 |
if document["score"] is not None:
|
69 |
st.markdown(f"**Score**: {document['score']:.3f}")
|
70 |
st.markdown("---")
|
71 |
|
72 |
|
73 |
+
def component_text_input(container, doc_id):
|
74 |
"""Draw the Text Input widget"""
|
75 |
with container:
|
76 |
texts = []
|
|
|
77 |
with st.expander("Enter documents"):
|
78 |
while True:
|
79 |
text = st.text_input(f"Document {doc_id}", key=doc_id)
|
80 |
if text != "":
|
81 |
+
texts.append({"text": text, "doc_id": doc_id})
|
82 |
doc_id += 1
|
83 |
st.markdown("---")
|
84 |
else:
|
85 |
break
|
86 |
+
corpus = [{"text": doc["text"], "id": doc["doc_id"]} for doc in texts]
|
87 |
+
return corpus, doc_id
|
|
|
|
|
88 |
|
89 |
|
90 |
+
def component_article_url(container, doc_id):
|
91 |
"""Draw the Article URL widget"""
|
92 |
with container:
|
93 |
urls = []
|
|
|
94 |
with st.expander("Enter URLs"):
|
95 |
while True:
|
96 |
url = st.text_input(f"URL {doc_id}", key=doc_id)
|
97 |
if url != "":
|
98 |
+
urls.append({"text": extract_text_from_url(url), "doc_id": doc_id})
|
99 |
doc_id += 1
|
100 |
st.markdown("---")
|
101 |
else:
|
|
|
103 |
|
104 |
for idx, doc in enumerate(urls):
|
105 |
with st.expander(f"Preview URL {idx}"):
|
106 |
+
st.write(doc["text"])
|
107 |
|
108 |
+
corpus = [{"text": doc["text"], "id": doc["doc_id"]} for doc in urls]
|
109 |
+
return corpus, doc_id
|
|
|
|
|
110 |
|
111 |
|
112 |
+
def component_file_input(container, doc_id):
|
113 |
"""Draw the extract text from file widget"""
|
114 |
with container:
|
115 |
files = []
|
|
|
116 |
with st.expander("Enter Files"):
|
117 |
while True:
|
118 |
file = st.file_uploader(
|
|
|
121 |
if file != None:
|
122 |
extracted_text = extract_text_from_file(file)
|
123 |
if extracted_text != None:
|
124 |
+
files.append({"text": extracted_text, "doc_id": doc_id})
|
125 |
doc_id += 1
|
126 |
st.markdown("---")
|
127 |
else:
|
|
|
131 |
|
132 |
for idx, doc in enumerate(files):
|
133 |
with st.expander(f"Preview File {idx}"):
|
134 |
+
st.write(doc["text"])
|
135 |
|
136 |
+
corpus = [{"text": doc["text"], "id": doc["doc_id"]} for doc in files]
|
137 |
+
return corpus, doc_id
|
|
|
|
interface/config.py
CHANGED
@@ -1,7 +1,11 @@
|
|
1 |
from interface.pages import page_landing_page, page_search, page_index
|
2 |
|
3 |
# Define default Session Variables over the whole session.
|
4 |
-
session_state_variables = {
|
|
|
|
|
|
|
|
|
5 |
|
6 |
# Define Pages for the demo
|
7 |
pages = {
|
|
|
1 |
from interface.pages import page_landing_page, page_search, page_index
|
2 |
|
3 |
# Define default Session Variables over the whole session.
|
4 |
+
session_state_variables = {
|
5 |
+
"pipeline": None,
|
6 |
+
"pipeline_func_parameters": [],
|
7 |
+
"doc_id": 0,
|
8 |
+
}
|
9 |
|
10 |
# Define Pages for the demo
|
11 |
pages = {
|
interface/pages.py
CHANGED
@@ -79,14 +79,17 @@ def page_index(container):
|
|
79 |
orientation="horizontal",
|
80 |
)
|
81 |
|
82 |
-
|
|
|
|
|
|
|
83 |
|
84 |
if len(corpus) > 0:
|
85 |
index_results = None
|
86 |
if st.button("Index"):
|
87 |
index_results = index(
|
88 |
-
corpus,
|
89 |
-
st.session_state["pipeline"]["index_pipeline"],
|
90 |
)
|
|
|
91 |
if index_results:
|
92 |
st.write(index_results)
|
|
|
79 |
orientation="horizontal",
|
80 |
)
|
81 |
|
82 |
+
clear_index = st.sidebar.checkbox("Clear Index", True)
|
83 |
+
|
84 |
+
doc_id = st.session_state["doc_id"]
|
85 |
+
corpus, doc_id = input_funcs[selected_input][0](container, doc_id)
|
86 |
|
87 |
if len(corpus) > 0:
|
88 |
index_results = None
|
89 |
if st.button("Index"):
|
90 |
index_results = index(
|
91 |
+
corpus, st.session_state["pipeline"]["index_pipeline"], clear_index
|
|
|
92 |
)
|
93 |
+
st.session_state["doc_id"] = doc_id
|
94 |
if index_results:
|
95 |
st.write(index_results)
|