File size: 6,584 Bytes
8d3914e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a577776
8d3914e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
67c37fc
8d3914e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ed4afb1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
import asyncio
import traceback

import gradio as gr

import config
from retrievers.kaggle import KaggleRetriever
from retrievers.huggingface import HuggingFaceRetriever
from retrievers.datagov import DataGovRetriever
from processing.normalizer import normalize_results
from processing.embedder import Embedder
from processing.query_intent import parse_query_intent
from processing.scorer import hard_filter, score_results
from processing.faiss_index import FAISSIndex
from ranking.llm_ranker import LLMRanker
from ranking.result_builder import build_result_cards

embedder = Embedder(model_name=config.EMBEDDING_MODEL)
ranker = LLMRanker(
    model_id=config.LLM_MODEL_ID)

async def _fetch_all(query: str) -> list[dict]:
    secrets = config.check_secrets()

    retrievers = [
        HuggingFaceRetriever(max_results=config.RESULTS_PER_SOURCE),
        DataGovRetriever(
            max_results=config.RESULTS_PER_SOURCE,
            timeout=config.RETRIEVER_TIMEOUT_SECONDS,
        ),
    ]
    if secrets["kaggle"]:
        retrievers.append(
            KaggleRetriever(
                username=config.KAGGLE_USERNAME,
                key=config.KAGGLE_KEY,
                max_results=config.RESULTS_PER_SOURCE,
            )
        )

    tasks = [r.fetch(query) for r in retrievers]
    results_per_lane = await asyncio.gather(*tasks, return_exceptions=True)

    merged: list[dict] = []
    for lane_result in results_per_lane:
        if isinstance(lane_result, Exception):
            print(f"[retriever error] {lane_result}")
            continue
        merged.extend(lane_result)

    return merged


def run_pipeline(query: str) -> tuple[str, list[dict]]:
    query = query.strip()
    if not query:
        return "Please enter a research question.", []

    try:
        intent = parse_query_intent(query)
        print(f"[intent] {intent}")

        raw_results = asyncio.run(_fetch_all(query))
        
        source_list = [res.get("source") for res in raw_results]
        source_counts = {src: source_list.count(src) for src in set(source_list)}
        print(f"DEBUG SOURCE COUNTS: {source_counts}")

        if not raw_results:
            return "No datasets found. Try rephrasing your question.", []

        normalized = normalize_results(raw_results)

        candidates, rejected = hard_filter(normalized, intent)

        if not candidates:
            candidates = normalized
            filter_warning = (
                f" Note: no datasets matched all constraints "
                f"({intent.summary()}), showing best available results."
            )
        else:
            filter_warning = ""

        query_embedding = embedder.embed_query(query)
        dataset_embeddings = embedder.embed_datasets(candidates)

        index = FAISSIndex()
        index.build(dataset_embeddings)
        top_indices = index.search(query_embedding, k=config.FAISS_TOP_K)
        faiss_candidates = [candidates[i] for i in top_indices]

        semantic_scores = {
            candidates[i]["name"]: float(score)
            for i, score in zip(top_indices, index.last_scores)
        }
        scored = score_results(
            datasets=faiss_candidates,
            query=query,
            intent=intent,
            semantic_scores=semantic_scores,
        )

        ranked = ranker.rank(
            query=query,
            candidates=scored[: config.LLM_INPUT_COUNT],
            intent_context=intent.context_signals,
            active_constraints=intent.hard_constraints,
        )

        cards = build_result_cards(ranked, top_n=config.DISPLAY_TOP_N)

        sources_used = len({c["source"] for c in normalized})
        n_filtered = len(rejected)
        filter_note = (
            f" ({n_filtered} filtered by query constraints)"
            if n_filtered and not filter_warning
            else ""
        )

        status = (
            f"Found {len(raw_results)} datasets across {sources_used} source(s). "
            f"Showing top {len(cards)} ranked by suitability."
            f"{filter_note}{filter_warning}"
        )
        return status, cards

    except Exception:
        traceback.print_exc()
        return "An error occurred. Check the logs for details.", []

def _format_cards_as_markdown(cards: list[dict]) -> str:
    if not cards:
        return ""

    lines = []
    for card in cards:
        source_badge = f"`{card['source']}`"
        score_pct = f"{card['relevance_score']:.0%}"

        lines.append(f"### {card['rank']}. {card['name']}  {source_badge}")
        lines.append(
            f"**Suitability score:** {score_pct}  |  "
            f"**Format:** {card.get('format', 'N/A')}  |  "
            f"**License:** {card.get('license', 'N/A')}  |  "
            f"**Updated:** {card.get('last_updated', 'N/A')}"
        )

        constraints = card.get("active_constraints")
        if constraints:
            c_str = ", ".join(f"`{k}={v}`" for k, v in constraints.items())
            lines.append(f"_Matched constraints: {c_str}_")

        lines.append(f"\n{card.get('suitability_notes', '')}")
        if card.get("url"):
            lines.append(f"\n[View dataset →]({card['url']})")
        lines.append("\n---")

    return "\n".join(lines)

def gradio_handler(query: str):
    status, cards = run_pipeline(query)
    return status, _format_cards_as_markdown(cards)

_secrets = config.check_secrets()
_missing = [k for k, v in _secrets.items() if not v]
if _missing:
    print(
        f"[config] Optional secrets not set: {', '.join(_missing)}. "
    )

with gr.Blocks(title="Dataset Recommender") as demo:
    gr.Markdown("## Dataset Recommender")
    gr.Markdown(
        "Describe your research question and get ranked, open dataset recommendations "
        "from Kaggle, Hugging Face and data.gov, simultaneously."
    )

    with gr.Row():
        query_box = gr.Textbox(
            label="Research question or problem description",
            placeholder="e.g. imbalanced dataset for fraud detection with labeled transactions",
            lines=3,
            scale=4,
        )
        submit_btn = gr.Button("Find datasets", variant="primary", scale=1)

    status_box = gr.Textbox(label="Status", interactive=False, lines=2)
    results_box = gr.Markdown(label="Results")

    submit_btn.click(
        fn=gradio_handler,
        inputs=[query_box],
        outputs=[status_box, results_box],
    )
    query_box.submit(
        fn=gradio_handler,
        inputs=[query_box],
        outputs=[status_box, results_box],
    )

if __name__ == "__main__":
    demo.launch()