ONES-RS Financial Analysis

# Install from wheel (enterprise version)
pip install ones_rs-1.1.0-cp311-cp311-win_amd64.whl

# Verify installation
python -c "from ones_rs.ones_rs import version; print(version())"

from ones_rs.ones_rs import OnesEngine, version
from pathlib import Path

# Initialize the ONES engine
engine = OnesEngine()

# Load the Loughran-McDonald financial lexicon
lexicon_path = Path("data/domain_lexicons/loughran_mcdonald_lexicon.json")
num_terms = engine.load_lexicon(str(lexicon_path))

print(f"ONES-RS Version: {version()}")
print(f"Lexicon loaded: {num_terms} financial terms")
print(f"Engine ready for analysis")

import pandas as pd
from pathlib import Path

# Load the AllAgree subset
data_path = Path("data/FinancialPhraseBank-v1.0/Sentences_AllAgree.txt")

sentences = []
labels = []

with open(data_path, 'r', encoding='latin-1') as f:
    for line in f:
        if '@' in line:
            parts = line.strip().rsplit('@', 1)
            if len(parts) == 2:
                sentences.append(parts[0])
                labels.append(parts[1])

df = pd.DataFrame({'sentence': sentences, 'label': labels})

print(f"Total sentences: {len(df)}")
print(f"Label distribution:")
print(df['label'].value_counts())

# Sample financial sentence
text = "Operating profit rose to EUR 13.1 mn from EUR 8.7 mn in 2007."

# Get sentiment classification
sentiment = engine.classify_sentiment(text)
valence = engine.calculate_valence(text)
domain = engine.detect_domain(text)

print(f"Text: {text}")
print(f"Sentiment: {sentiment}")
print(f"Valence Score: {valence:.4f}")
print(f"Detected Domain: {domain}")

import time

# Get all sentences
all_sentences = df['sentence'].tolist()

# Batch analysis with automatic parallelization
start_time = time.time()
results = engine.analyze_batch_auto(all_sentences)
elapsed = time.time() - start_time

# Results format: (index, domain, sentiment, valence)
df['ones_sentiment'] = [r[2] for r in results]
df['ones_valence'] = [r[3] for r in results]

# Calculate throughput
throughput = len(all_sentences) / elapsed

print(f"Processed {len(all_sentences):,} texts in {elapsed:.4f} seconds")
print(f"Throughput: {throughput:,.0f} texts/second")

from sklearn.metrics import accuracy_score, classification_report

# Calculate overall accuracy
accuracy = accuracy_score(df['label'], df['ones_sentiment'])

print(f"Overall Accuracy: {accuracy:.2%}")
print()
print("Classification Report:")
print(classification_report(df['label'], df['ones_sentiment']))

# Define custom financial labels
custom_labels = [
    ("bullish", "strong growth, positive outlook, exceeds expectations, profitable, increase"),
    ("bearish", "decline, loss, negative outlook, underperform, weak, decrease"),
    ("neutral", "stable, unchanged, maintains, as expected")
]

# Classify sample texts
test_texts = [
    "Revenue grew 25% and exceeded all analyst expectations.",
    "The company reported a significant quarterly loss.",
    "Operations continued as normal with stable margins."
]

for text in test_texts:
    label, score = engine.classify_to_label(text, custom_labels)
    print(f"[{label:8}] ({score:.3f}) {text}")

# Texts to cluster
texts = [
    "Revenue increased significantly",
    "Revenue grew strongly",
    "Sales declined sharply",
    "Sales dropped significantly",
    "Earnings were stable"
]

# Group by similarity (threshold 0.2)
groups = engine.group_by_similarity(texts, 0.2)

print("Clustering Results:")
for text, cluster in zip(texts, groups):
    print(f"  Cluster {cluster}: {text}")