optz the data loading
Browse files
app.py
CHANGED
@@ -3,57 +3,64 @@ import pandas as pd
|
|
3 |
from datasets import load_dataset
|
4 |
import jiwer
|
5 |
import numpy as np
|
|
|
6 |
|
7 |
-
#
|
|
|
8 |
def load_data():
|
9 |
-
|
10 |
-
return dataset
|
11 |
|
12 |
# Calculate WER for a group of examples
|
13 |
def calculate_wer(examples):
|
14 |
if not examples:
|
15 |
return 0.0
|
16 |
|
17 |
-
|
|
|
|
|
|
|
18 |
|
19 |
-
for ex in examples:
|
20 |
-
# Get transcription and input1 fields
|
21 |
-
transcription = ex.get("transcription")
|
22 |
-
input1 = ex.get("input1")
|
23 |
-
|
24 |
-
# Only include examples where both fields exist and are not empty
|
25 |
-
if transcription and input1:
|
26 |
-
valid_pairs.append((transcription.strip(), input1.strip()))
|
27 |
-
|
28 |
-
# If no valid pairs were found, return NaN
|
29 |
if not valid_pairs:
|
30 |
return np.nan
|
31 |
|
32 |
-
#
|
33 |
-
references =
|
34 |
-
hypotheses = [pair[1] for pair in valid_pairs]
|
35 |
|
36 |
# Calculate WER
|
37 |
-
|
38 |
-
return wer
|
39 |
|
40 |
# Get WER metrics by source and split
|
41 |
def get_wer_metrics(dataset):
|
42 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
|
|
48 |
|
49 |
-
#
|
|
|
|
|
|
|
|
|
50 |
for source in all_sources:
|
51 |
-
train_examples =
|
52 |
-
|
53 |
-
train_wer = calculate_wer(train_examples) if train_count > 0 else np.nan
|
54 |
|
55 |
-
|
56 |
test_count = len(test_examples)
|
|
|
|
|
57 |
test_wer = calculate_wer(test_examples) if test_count > 0 else np.nan
|
58 |
|
59 |
results.append({
|
@@ -64,7 +71,7 @@ def get_wer_metrics(dataset):
|
|
64 |
"Test WER": test_wer
|
65 |
})
|
66 |
|
67 |
-
#
|
68 |
train_wer = calculate_wer(dataset["train"])
|
69 |
test_wer = calculate_wer(dataset["test"])
|
70 |
|
@@ -80,8 +87,16 @@ def get_wer_metrics(dataset):
|
|
80 |
|
81 |
# Format the dataframe for display
|
82 |
def format_dataframe(df):
|
83 |
-
|
84 |
-
df
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
85 |
return df
|
86 |
|
87 |
# Main function to create the leaderboard
|
@@ -89,8 +104,7 @@ def create_leaderboard():
|
|
89 |
try:
|
90 |
dataset = load_data()
|
91 |
metrics_df = get_wer_metrics(dataset)
|
92 |
-
|
93 |
-
return formatted_df
|
94 |
except Exception as e:
|
95 |
return pd.DataFrame({"Error": [str(e)]})
|
96 |
|
|
|
3 |
from datasets import load_dataset
|
4 |
import jiwer
|
5 |
import numpy as np
|
6 |
+
from functools import lru_cache
|
7 |
|
8 |
+
# Cache the dataset loading to avoid reloading on refresh
|
9 |
+
@lru_cache(maxsize=1)
|
10 |
def load_data():
|
11 |
+
return load_dataset("GenSEC-LLM/SLT-Task1-Post-ASR-Text-Correction")
|
|
|
12 |
|
13 |
# Calculate WER for a group of examples
|
14 |
def calculate_wer(examples):
|
15 |
if not examples:
|
16 |
return 0.0
|
17 |
|
18 |
+
# Filter valid examples in a single pass
|
19 |
+
valid_pairs = [(ex.get("transcription", "").strip(), ex.get("input1", "").strip())
|
20 |
+
for ex in examples
|
21 |
+
if ex.get("transcription") and ex.get("input1")]
|
22 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
if not valid_pairs:
|
24 |
return np.nan
|
25 |
|
26 |
+
# Unzip the pairs in one operation
|
27 |
+
references, hypotheses = zip(*valid_pairs) if valid_pairs else ([], [])
|
|
|
28 |
|
29 |
# Calculate WER
|
30 |
+
return jiwer.wer(references, hypotheses)
|
|
|
31 |
|
32 |
# Get WER metrics by source and split
|
33 |
def get_wer_metrics(dataset):
|
34 |
+
# Pre-process the data to avoid repeated filtering
|
35 |
+
train_by_source = {}
|
36 |
+
test_by_source = {}
|
37 |
+
|
38 |
+
# Group examples by source in a single pass for each split
|
39 |
+
for ex in dataset["train"]:
|
40 |
+
source = ex["source"]
|
41 |
+
if source not in train_by_source:
|
42 |
+
train_by_source[source] = []
|
43 |
+
train_by_source[source].append(ex)
|
44 |
|
45 |
+
for ex in dataset["test"]:
|
46 |
+
source = ex["source"]
|
47 |
+
if source not in test_by_source:
|
48 |
+
test_by_source[source] = []
|
49 |
+
test_by_source[source].append(ex)
|
50 |
|
51 |
+
# Get all unique sources
|
52 |
+
all_sources = sorted(set(train_by_source.keys()) | set(test_by_source.keys()))
|
53 |
+
|
54 |
+
# Calculate metrics for each source
|
55 |
+
results = []
|
56 |
for source in all_sources:
|
57 |
+
train_examples = train_by_source.get(source, [])
|
58 |
+
test_examples = test_by_source.get(source, [])
|
|
|
59 |
|
60 |
+
train_count = len(train_examples)
|
61 |
test_count = len(test_examples)
|
62 |
+
|
63 |
+
train_wer = calculate_wer(train_examples) if train_count > 0 else np.nan
|
64 |
test_wer = calculate_wer(test_examples) if test_count > 0 else np.nan
|
65 |
|
66 |
results.append({
|
|
|
71 |
"Test WER": test_wer
|
72 |
})
|
73 |
|
74 |
+
# Calculate overall metrics once
|
75 |
train_wer = calculate_wer(dataset["train"])
|
76 |
test_wer = calculate_wer(dataset["test"])
|
77 |
|
|
|
87 |
|
88 |
# Format the dataframe for display
|
89 |
def format_dataframe(df):
|
90 |
+
# Use vectorized operations instead of apply
|
91 |
+
df = df.copy()
|
92 |
+
mask = df["Train WER"].notna()
|
93 |
+
df.loc[mask, "Train WER"] = df.loc[mask, "Train WER"].map(lambda x: f"{x:.4f}")
|
94 |
+
df.loc[~mask, "Train WER"] = "N/A"
|
95 |
+
|
96 |
+
mask = df["Test WER"].notna()
|
97 |
+
df.loc[mask, "Test WER"] = df.loc[mask, "Test WER"].map(lambda x: f"{x:.4f}")
|
98 |
+
df.loc[~mask, "Test WER"] = "N/A"
|
99 |
+
|
100 |
return df
|
101 |
|
102 |
# Main function to create the leaderboard
|
|
|
104 |
try:
|
105 |
dataset = load_data()
|
106 |
metrics_df = get_wer_metrics(dataset)
|
107 |
+
return format_dataframe(metrics_df)
|
|
|
108 |
except Exception as e:
|
109 |
return pd.DataFrame({"Error": [str(e)]})
|
110 |
|