def run_workflow(jsonl_path: str,
doc_column: str = None,
reference_column: str = None,
summary_columns: List[str] = None,
bert_aligner_threshold: float = 0.5,
bert_aligner_top_k: int = 3,
embedding_aligner_threshold: float = 0.5,
embedding_aligner_top_k: int = 3,
processed_dataset_path: str = None,
n_samples: int = None):
if not jsonl_path:
raise ValueError("'jsonl_path' is required")
if not processed_dataset_path:
raise ValueError("Please specify a path to save the dataset.")
# Load the dataset
dataset = DataPanel.from_jsonl(jsonl_path)
if doc_column is None:
# Assume `doc_column` is called "document"
doc_column = 'document'
assert doc_column in dataset.columns, \
f"`doc_column={doc_column}` is not a column in datapanel."
print("Assuming `doc_column` is called 'document'.")
if reference_column is None:
# Assume `reference_column` is called "summary:reference"
reference_column = 'summary:reference'
print("Assuming `reference_column` is called 'summary:reference'.")
if reference_column not in dataset.columns:
print("No reference summary loaded")
reference_column = None
if summary_columns is None or len(summary_columns) == 0:
# Assume `summary_columns` are prefixed by "summary:"
summary_columns = []
for col in dataset.columns:
if col.startswith("summary:") and col != "summary:reference":
summary_columns.append(col)
print(
f"Reading summary columns from datapanel. Found {summary_columns}."
)
if len(summary_columns) == 0 and reference_column is None:
raise ValueError("At least one summary is required")
# Restrict to the first `n_samples`
if n_samples:
print(f"Restricting to {n_samples} samples.")
dataset = dataset.head(n_samples)
print("size of dataset:", len(dataset))
# Combine the text columns into one list
text_columns = [doc_column] + ([reference_column] if reference_column else
[]) + summary_columns
# Preprocessing all the text columns
print("Preprocessing text columns")
dataset = dataset.update(
lambda x: {
f'preprocessed_{k}': x[k] if args.no_clean else clean_text(x[k])
for k in text_columns
})
# Run the Spacy pipeline on all preprocessed text columns
nlp = load_nlp()
nlp.add_pipe('sentencizer', before="parser")
print("Running spacy processing")
for col in text_columns:
dataset.add_column(
f'spacy:{col}',
SpacyColumn.from_docs(nlp.pipe(dataset[f'preprocessed_{col}'])))
# Run the 3 align pipelines
bert_aligner = BertscoreAligner(
threshold=bert_aligner_threshold,
top_k=bert_aligner_top_k,
)
embedding_aligner = StaticEmbeddingAligner(
threshold=embedding_aligner_threshold,
top_k=embedding_aligner_top_k,
)
ngram_aligner = NGramAligner()
dataset = _run_aligners(
dataset=dataset,
aligners=[bert_aligner, embedding_aligner, ngram_aligner],
doc_column=f'spacy:{doc_column}',
reference_column=f'spacy:{reference_column}'
if reference_column else None,
summary_columns=[f'spacy:{col}' for col in summary_columns],
)
# Save the dataset
dataset.write(processed_dataset_path)
return dataset
def __init__(
self,
spacy,
):
super(NGramAlignerCap, self).__init__(aligner=NGramAligner(),
spacy=spacy)
def show_main(example):
# Get user input
semantic_sim_type = st.sidebar.radio(
"Semantic similarity type:",
["Contextual embedding", "Static embedding"])
semantic_sim_threshold = st.sidebar.slider(
"Semantic similarity threshold:",
min_value=MIN_SEMANTIC_SIM_THRESHOLD,
max_value=1.0,
step=0.1,
value=0.2,
)
semantic_sim_top_k = st.sidebar.slider(
"Semantic similarity top-k:",
min_value=1,
max_value=MAX_SEMANTIC_SIM_TOP_K,
step=1,
value=10,
)
document, summaries = select_comparison(example)
layout = st.sidebar.radio("Layout:", ["Vertical", "Horizontal"]).lower()
# if layout == "horizontal":
# scroll = st.sidebar.checkbox(label="Scroll sections", value=True)
# else:
scroll = True
grey_stopwords = st.sidebar.checkbox(label="Grey out stopwords",
value=True)
# Gather data
try:
lexical_alignments = [
NGramAlignerCap.decode(example.data[Identifier(
NGramAlignerCap.__name__)(columns=[
f'preprocessed_{document._.column}',
f'preprocessed_{summary._.column}',
])])[0] for summary in summaries
]
lexical_alignments = [{
k: [(pair[0], int(pair[1])) for pair in v]
for k, v in d.items()
} for d in lexical_alignments]
except KeyError:
lexical_alignments = NGramAligner().align(document, summaries)
if semantic_sim_type == "Static embedding":
try:
semantic_alignments = [
StaticEmbeddingAlignerCap.decode(example.data[Identifier(
StaticEmbeddingAlignerCap.__name__)(
threshold=MIN_SEMANTIC_SIM_THRESHOLD,
top_k=MAX_SEMANTIC_SIM_TOP_K,
columns=[
f'preprocessed_{document._.column}',
f'preprocessed_{summary._.column}',
])])[0] for summary in summaries
]
except KeyError:
semantic_alignments = StaticEmbeddingAligner(
semantic_sim_threshold,
semantic_sim_top_k).align(document, summaries)
else:
semantic_alignments = [
filter_alignment(alignment, semantic_sim_threshold,
semantic_sim_top_k)
for alignment in semantic_alignments
]
else:
try:
semantic_alignments = [
BertscoreAlignerCap.decode(example.data[Identifier(
BertscoreAlignerCap.__name__)(
threshold=MIN_SEMANTIC_SIM_THRESHOLD,
top_k=MAX_SEMANTIC_SIM_TOP_K,
columns=[
f'preprocessed_{document._.column}',
f'preprocessed_{summary._.column}',
])])[0] for summary in summaries
]
except KeyError:
semantic_alignments = BertscoreAligner(semantic_sim_threshold,
semantic_sim_top_k).align(
document, summaries)
else:
semantic_alignments = [
filter_alignment(alignment, semantic_sim_threshold,
semantic_sim_top_k)
for alignment in semantic_alignments
]
show_html(*main_view(document, summaries, semantic_alignments,
lexical_alignments, layout, scroll, grey_stopwords),
height=850)
def show_main(example):
# Get user input
semantic_sim_type = st.sidebar.radio(
"Semantic similarity type:",
["Contextual embedding", "Static embedding"])
semantic_sim_threshold = st.sidebar.slider(
"Semantic similarity threshold:",
min_value=MIN_SEMANTIC_SIM_THRESHOLD,
max_value=1.0,
step=0.1,
value=0.2,
)
semantic_sim_top_k = st.sidebar.slider(
"Semantic similarity top-k:",
min_value=1,
max_value=MAX_SEMANTIC_SIM_TOP_K,
step=1,
value=10,
)
document, summaries = select_comparison(example)
layout = st.sidebar.radio("Layout:", ["Vertical", "Horizontal"]).lower()
scroll = True
gray_out_stopwords = st.sidebar.checkbox(label="Gray out stopwords",
value=True)
# Gather data
try:
lexical_alignments = [
example.data[
f'{NGramAligner.__name__}:spacy:{document._.column}:spacy:{summary._.column}']
for summary in summaries
]
except KeyError:
lexical_alignments = NGramAligner().align(document, summaries)
if semantic_sim_type == "Static embedding":
try:
semantic_alignments = [
example.data[
f'{StaticEmbeddingAligner.__name__}:spacy:{document._.column}:spacy:{summary._.column}']
for summary in summaries
]
except KeyError:
semantic_alignments = StaticEmbeddingAligner(
semantic_sim_threshold,
semantic_sim_top_k).align(document, summaries)
else:
try:
semantic_alignments = [
example.data[
f'{BertscoreAligner.__name__}:spacy:{document._.column}:spacy:{summary._.column}']
for summary in summaries
]
except KeyError:
semantic_alignments = BertscoreAligner(semantic_sim_threshold,
semantic_sim_top_k).align(
document, summaries)
MainView(
document,
summaries,
semantic_alignments,
lexical_alignments,
layout,
scroll,
gray_out_stopwords,
).show(height=720)