vishnu-coder commited on
Commit
055039a
·
2 Parent(s): eab22563473e65

Merge branch 'main' into codex/enhance-and-deploy-project-for-interview-c6c16v

Browse files
Files changed (2) hide show
  1. README.md +108 -104
  2. app/app.py +27 -52
README.md CHANGED
@@ -1,138 +1,137 @@
1
- # 💼 Deloitte-Ready Twitter Sentiment Intelligence
2
 
3
- A production-grade sentiment analysis accelerator designed to showcase the end-to-end delivery skills expected from a Deloitte India Oracle Analyst. The solution covers data ingestion, preprocessing, model training, governance, deployment readiness, and professional storytelling while retaining the lightweight learning experience from the original project.
 
4
 
5
- ## 🧰 Technology stack & analytics capabilities
6
 
7
- - Python, Pandas, Scikit-learn, and NLTK for rapid experimentation and production-ready sentiment pipelines.
8
- - TF-IDF vectorization paired with multiclass logistic regression for explainable predictions across positive, neutral, and negative classes.
9
- - Streamlit web UI (`app/app.py`) that surfaces class probabilities and narrative insights for stakeholders.
10
- - Notebook-driven exploration (`notebooks/`) to trace the lineage of the trained pipeline and retrain artefacts when required.
11
 
 
12
 
13
- ## 🧭 Why this project stands out
14
 
15
- - **Oracle ecosystem alignment** – configurable ingestion from Oracle Autonomous Database, OCI deployment guide, and analytics workflows tailored for enterprise reporting.
16
- - **Consulting-grade engineering** – modular Python package, logging, configuration management, and CI/CD automation.
17
- - **Business storytelling** – dashboards, metrics, and documentation that translate technical outcomes into measurable value for stakeholders.
18
 
19
- ## 🔗 Live resources & quick links
20
 
21
- | Asset | Purpose | Link |
22
- | --- | --- | --- |
23
- | **Live Streamlit demo** | Shareable dashboard URL for interviews (update after deployment) | [`https://<your-app>.streamlit.app`](https://<your-app>.streamlit.app) |
24
- | **Architecture blueprint** | Deep dive on data, model, and DevOps layers | [`docs/architecture.md`](docs/architecture.md) |
25
- | **OCI deployment playbook** | Container-first rollout with Oracle services | [`deployment/oracle_cloud.md`](deployment/oracle_cloud.md) |
26
- | **Streamlit Community Cloud guide** | Launch a public URL in minutes | [`deployment/streamlit_cloud.md`](deployment/streamlit_cloud.md) |
27
- | **Vercel redirect playbook** | Turn your Vercel domain into a Streamlit launcher | [`deployment/vercel_redirect.md`](deployment/vercel_redirect.md) |
28
- | **GitHub connection & live check** | Step-by-step repo linking plus URL validation | [`deployment/github_live_validation.md`](deployment/github_live_validation.md) |
29
- | **Training data sample** | Jump-start retraining conversations | [`data/twitter_training.csv`](data/twitter_training.csv) |
30
 
31
- > ✅ Once you publish the Streamlit app, replace the placeholder domain above. The link will then be accessible from any device (mobile, tablet, or laptop) and can be embedded directly in your resume or interview slides.
32
 
33
- ## 🏗️ Architecture overview
34
 
35
- ```mermaid
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
36
  flowchart LR
37
- subgraph Data Layer
38
  A[Oracle Autonomous DB] -- optional --> B[(CSV Data)]
39
  end
40
- subgraph Processing Layer
41
  B --> C[Data Loader]
42
  C --> D[Text Preprocessor]
43
  D --> E[Scikit-learn Pipeline]
44
  E --> F[Model Metrics]
45
  end
46
- subgraph Experience Layer
47
  E --> G[Streamlit Dashboard]
48
  E --> H[CLI Automation]
49
  F --> G
50
  F --> I[Reporting / GitHub Pages]
51
  end
52
- subgraph DevOps Layer
53
  J[GitHub Actions CI]
54
  J --> E
55
  J --> G
56
  end
57
- ```
58
-
59
- ## 📁 Repository structure
60
-
61
- | Path | Description |
62
- | --- | --- |
63
- | `app/` | Streamlit application for stakeholder demos |
64
- | `artifacts/` | Generated model artifacts (gitignored; create via `scripts/train.py` before first run) |
65
- | `config/` | Centralised YAML configuration controlling ingestion, preprocessing, and deployment toggles (stored in JSON-compatible YAML for portability) |
66
- | `data/` | Sample labelled tweets for local experimentation |
67
- | `deployment/` | Cloud deployment runbooks (Oracle Cloud Infrastructure, Azure, AWS) |
68
- | `docs/` | Architecture diagrams, KPI catalogue, and interview collateral |
69
- | `scripts/` | Automation scripts for training and inference |
70
- | `src/twitter_sentiment/` | Reusable Python package powering the pipeline |
71
- | `tests/` | Pytest-based unit tests covering preprocessing and inference |
72
- | `.github/workflows/` | CI workflow executing tests and lightweight linting |
73
-
74
- ## 🚀 Quick start
75
-
76
- ```bash
77
- # 1. Create a virtual environment (recommended)
78
  python -m venv .venv
79
  source .venv/bin/activate # Windows: .venv\Scripts\activate
80
 
81
- # 2. Install dependencies
82
  pip install --upgrade pip
83
  pip install -r requirements.txt
84
 
85
- # 3. Train the pipeline to generate fresh artifacts
86
  python scripts/train.py
87
 
88
- # 4. Launch the Deloitte storytelling dashboard
89
  streamlit run app/app.py
90
- ```
91
 
92
- > **Note:** Model artifacts are deliberately excluded from source control. Run `python scripts/train.py` (or `make train`) any time you clone the repo or reset the workspace so that `artifacts/sentiment_pipeline.joblib` is available for the app, CLI, and tests.
93
 
94
- Prefer a lightweight launch without the Deloitte extensions? Deploy the original Streamlit interface directly via `streamlit run app/app.py` or reuse the serialized pipeline from the `artifacts/` directory once generated.
 
 
 
 
 
 
 
95
 
96
- ## 🧪 Quality gates
97
 
98
- | Check | Command |
99
- | --- | --- |
100
- | Unit tests | `pytest -q` |
101
- | Package sanity | `python -m compileall src` |
102
- | Model retrain | `python scripts/train.py` |
103
 
104
- GitHub Actions automatically executes tests and sanity checks on every push / PR against `main`.
105
 
106
- ## 🧩 Feature highlights
107
 
108
- - Config-driven ingestion toggling between CSV and Oracle Autonomous Database.
109
- - Reusable preprocessing component with whitespace, URL, mention, and punctuation handling.
110
- - Scikit-learn pipeline persisted with evaluation metrics for auditability.
111
- - Streamlit dashboard with business-friendly narrative, download-ready metrics, and governance tab.
112
- - CLI utilities for automation workflows (`scripts/train.py`, `scripts/predict.py`).
113
- - Unit tests and CI pipeline to demonstrate engineering rigor.
114
 
115
- ## ☁️ Deploy to the cloud
116
 
117
- A detailed, step-by-step OCI deployment guide lives at [`deployment/oracle_cloud.md`](deployment/oracle_cloud.md). Highlights include:
118
 
119
- 1. Containerising the Streamlit app with OCI Container Instances.
120
- 2. Automating retraining via OCI Data Science jobs or GitHub Actions.
121
- 3. Wiring up Autonomous Database sources using wallet-based connectivity.
122
 
123
- Need a fast public URL you can open from any phone? Follow the Streamlit Community Cloud steps in [`deployment/streamlit_cloud.md`](deployment/streamlit_cloud.md) and drop the generated link into the **Live resources** table above.
124
 
125
- Once your link is live, update `live_app.streamlit_url` in [`config/settings.yaml`](config/settings.yaml) and run `make verify-live` (or invoke [`scripts/check_live_app.py`](scripts/check_live_app.py)) with no arguments. The command now reads the config value and confirms the dashboard responds with `200 OK` before interviews.
 
126
 
127
- Need a vanity Vercel link for QR codes? Generate the redirect HTML via [`scripts/update_vercel_redirect.py`](scripts/update_vercel_redirect.py) and follow [`deployment/vercel_redirect.md`](deployment/vercel_redirect.md) to publish it.
128
 
129
- Additional notes for Azure Web Apps and AWS App Runner are provided for multi-cloud discussions.
130
 
131
- ## 🗃️ Oracle database integration
132
 
133
- Update [`config/settings.yaml`](config/settings.yaml) to toggle `oracle_integration.enabled` and supply wallet credentials. The data loader will then pull training data directly from the Autonomous Database using `oracledb`.
 
134
 
135
- ```yaml
136
  oracle_integration:
137
  enabled: true
138
  wallet_location: /path/to/wallet
@@ -142,40 +141,45 @@ oracle_integration:
142
  SELECT text, sentiment
143
  FROM analytics.twitter_training_data
144
  WHERE created_at >= SYSDATE - 30
145
- ```
146
 
147
- ## 📊 Storytelling in interviews
 
 
 
 
 
 
 
 
 
 
 
 
148
 
149
- 1. **Business impact** – emphasise how proactive sentiment tracking improves customer retention and campaign ROI.
150
- 2. **Oracle expertise** – discuss wallet-based connectivity, SQL data modelling, and how OCI services plug into the pipeline.
151
- 3. **Engineering excellence** – highlight modular package design, automated tests, and CI/CD.
152
- 4. **Consulting mindset** – walk through the Streamlit dashboard as an executive-ready deliverable with actionable insights.
153
 
154
- ## 📦 Publishing on GitHub
155
 
156
- - Use feature branches for enhancements (`feature/oracle-ingestion`, `chore/ci-updates`).
157
- - Raise Pull Requests with CI status checks and screenshots of the Streamlit UI.
158
- - Leverage GitHub Projects to track backlog items (e.g., drift monitoring, Oracle APEX reporting).
159
- - Tag releases (e.g., `v1.0.0`) once the model is retrained and deployment is live.
160
 
161
- ## 📣 Share the live demo
162
 
163
- - Deploy the Streamlit container to OCI or Streamlit Community Cloud.
164
- - Capture dashboard screenshots (`docs/screenshots/`) for the README and interview slide deck.
165
- - Prepare a 3-minute walkthrough focusing on client scenario, data sources, and automation roadmap.
166
 
167
- ## 🤝 Contributing
168
 
169
- Pull requests are welcome! Please open an issue describing the proposed enhancement or bug fix before submitting large changes.
170
 
171
- ---
172
 
173
- Made with consulting rigour to help you shine in the Deloitte India Oracle Analyst interview.
174
 
175
- ## 🧠 Training artifacts
176
 
177
- This repository intentionally keeps generated artifacts out of Git history, but once you run the training script you will find the following in `artifacts/` (formerly committed under `app/`):
178
 
179
- - `sentiment_pipeline.joblib` – end-to-end Scikit-learn pipeline containing TF-IDF vectorizer and logistic regression classifier.
180
 
181
- Legacy artifacts (`app/logistic_model.pkl`, `app/tfidf_vectorizer.pkl`, `sentiment_model.pkl`) produced by the original notebook are still compatible with the refreshed Streamlit UI. Regenerate them via the notebooks in `notebooks/` if you need backwards-compatible demos or comparisons.
 
 
 
1
+ 💼 Deloitte-Ready Twitter Sentiment Intelligence
2
 
3
+ A production-grade sentiment analysis accelerator designed to showcase the end-to-end delivery skills expected from a Deloitte India Oracle Analyst.
4
+ The solution spans data ingestion, preprocessing, model training, governance, deployment readiness, and business storytelling — all within a lightweight, reproducible ML pipeline.
5
 
6
+ 🧰 Technology Stack & Analytics Capabilities
7
 
8
+ Python, Pandas, Scikit-learn, NLTK – rapid experimentation & production-ready sentiment pipelines
 
 
 
9
 
10
+ TF-IDF Vectorization + Multiclass Logistic Regression – explainable predictions across positive, neutral, negative classes
11
 
12
+ Streamlit UI (app/app.py) – displays class probabilities & narrative insights for stakeholders
13
 
14
+ Notebook-driven exploration (notebooks/) – retraining lineage and model interpretability
 
 
15
 
16
+ Container-ready CI/CD pipeline – automates tests, linting, and retraining
17
 
18
+ 🧭 Why This Project Stands Out
 
 
 
 
 
 
 
 
19
 
20
+ Oracle ecosystem alignment – optional ingestion from Oracle Autonomous Database with OCI deployment playbooks
21
 
22
+ Consulting-grade engineering – modular Python package, logging, configuration management, GitHub Actions CI
23
 
24
+ Business storytelling – dashboards, KPIs, and executive-friendly insights translating data into measurable ROI
25
+
26
+ 🔗 Live Resources & Quick Links
27
+ Asset Purpose Link
28
+ Live Streamlit Demo Interview-ready interactive dashboard https://<your-app>.streamlit.app
29
+
30
+ Architecture Blueprint Deep dive on data, model & DevOps layers docs/architecture.md
31
+
32
+ OCI Deployment Playbook Containerized rollout on Oracle Cloud deployment/oracle_cloud.md
33
+
34
+ Streamlit Cloud Guide Launch public URL in minutes deployment/streamlit_cloud.md
35
+
36
+ Vercel Redirect Playbook Vanity domain / QR-friendly link deployment/vercel_redirect.md
37
+
38
+ GitHub Live Validation Verify live deployment health deployment/github_live_validation.md
39
+
40
+ Training Data Sample Quick-start dataset for retraining data/twitter_training.csv
41
+
42
+ ✅ After publishing, replace the placeholder URL above and embed it directly in your resume / interview slides.
43
+
44
+ 🏗️ Architecture Overview
45
  flowchart LR
46
+ subgraph Data_Layer
47
  A[Oracle Autonomous DB] -- optional --> B[(CSV Data)]
48
  end
49
+ subgraph Processing_Layer
50
  B --> C[Data Loader]
51
  C --> D[Text Preprocessor]
52
  D --> E[Scikit-learn Pipeline]
53
  E --> F[Model Metrics]
54
  end
55
+ subgraph Experience_Layer
56
  E --> G[Streamlit Dashboard]
57
  E --> H[CLI Automation]
58
  F --> G
59
  F --> I[Reporting / GitHub Pages]
60
  end
61
+ subgraph DevOps_Layer
62
  J[GitHub Actions CI]
63
  J --> E
64
  J --> G
65
  end
66
+
67
+ 📁 Repository Structure
68
+ Path Description
69
+ app/ Streamlit app for demos
70
+ artifacts/ Generated model artifacts (gitignored; created via scripts/train.py)
71
+ config/ Central YAML settings controlling ingestion & deployment toggles
72
+ data/ Sample labelled tweets
73
+ deployment/ Multi-cloud deployment guides
74
+ docs/ Architecture diagrams & KPI catalogue
75
+ scripts/ Automation scripts for training & inference
76
+ src/twitter_sentiment/ Core reusable Python package
77
+ tests/ Pytest-based unit tests
78
+ .github/workflows/ CI pipeline (lint + tests)
79
+ 🚀 Quick Start
80
+ # 1️⃣ Create virtual environment
 
 
 
 
 
 
81
  python -m venv .venv
82
  source .venv/bin/activate # Windows: .venv\Scripts\activate
83
 
84
+ # 2️⃣ Install dependencies
85
  pip install --upgrade pip
86
  pip install -r requirements.txt
87
 
88
+ # 3️⃣ Train pipeline & generate artifacts
89
  python scripts/train.py
90
 
91
+ # 4️⃣ Launch Deloitte Storytelling Dashboard
92
  streamlit run app/app.py
 
93
 
 
94
 
95
+ Note: Artifacts are excluded from source control.
96
+ Run python scripts/train.py whenever cloning the repo to recreate artifacts/sentiment_pipeline.joblib.
97
+
98
+ 🧪 Quality Gates
99
+ Check Command
100
+ Unit tests pytest -q
101
+ Package compile check python -m compileall src
102
+ Model retrain python scripts/train.py
103
 
104
+ CI runs automatically on every push or PR to main.
105
 
106
+ 🧩 Feature Highlights
 
 
 
 
107
 
108
+ Config-driven ingestion switch (CSV ⇄ Oracle DB)
109
 
110
+ Reusable text-cleaning module (URL, mention, punctuation handling)
111
 
112
+ Auditable Scikit-learn pipeline with persisted metrics
 
 
 
 
 
113
 
114
+ Streamlit dashboard → probabilities + governance tab
115
 
116
+ CLI utilities for automated workflows
117
 
118
+ Pytest suite + GitHub Actions CI/CD
 
 
119
 
120
+ ☁️ Cloud Deployment
121
 
122
+ See deployment/oracle_cloud.md
123
+ for the Oracle Cloud Infrastructure (OCI) guide:
124
 
125
+ Containerize Streamlit app using OCI Container Instances
126
 
127
+ Automate retraining via OCI Data Science jobs / GitHub Actions
128
 
129
+ Connect to Autonomous DB with wallet credentials
130
 
131
+ Or follow deployment/streamlit_cloud.md
132
+ for a free Streamlit Cloud public link—perfect for interviews.
133
 
134
+ 🗃️ Oracle Database Integration (Optional)
135
  oracle_integration:
136
  enabled: true
137
  wallet_location: /path/to/wallet
 
141
  SELECT text, sentiment
142
  FROM analytics.twitter_training_data
143
  WHERE created_at >= SYSDATE - 30
 
144
 
145
+ 📊 Storytelling in Interviews
146
+
147
+ Business Impact – sentiment tracking → improved ROI & retention
148
+
149
+ Oracle Expertise – wallet connectivity, SQL modeling, OCI integration
150
+
151
+ Engineering Rigor – modular design, testing, CI/CD
152
+
153
+ Consulting Mindset – executive-ready Streamlit deliverable
154
+
155
+ 📦 GitHub Best Practices
156
+
157
+ Use feature branches (e.g., feature/oracle-ingestion)
158
 
159
+ Raise PRs with CI status + screenshots
 
 
 
160
 
161
+ Track backlog via GitHub Projects
162
 
163
+ Tag releases (e.g., v1.0.0) post-deployment
 
 
 
164
 
165
+ 🧠 Training Artifacts
166
 
167
+ Generated after running scripts/train.py:
 
 
168
 
169
+ sentiment_pipeline.joblib – Full TF-IDF + Logistic Regression pipeline
170
 
171
+ Legacy (compatible) files:
172
 
173
+ app/logistic_model.pkl
174
 
175
+ app/tfidf_vectorizer.pkl
176
 
177
+ sentiment_model.pkl
178
 
179
+ Use these for quick demos or backward comparisons.
180
 
181
+ 🤝 Contributing
182
 
183
+ Pull requests welcome!
184
+ Please open an issue describing any enhancement or bugfix before major changes.
185
+
app/app.py CHANGED
@@ -1,15 +1,18 @@
1
- """Streamlit front-end for the Deloitte-ready Twitter sentiment analyser."""
2
 
3
  from __future__ import annotations
4
 
5
  import json
6
- from pathlib import Path
7
  import sys
 
8
  from typing import Dict
9
 
10
  import pandas as pd
11
  import streamlit as st
12
 
 
 
 
13
  ROOT = Path(__file__).resolve().parents[1]
14
  SRC_PATH = ROOT / "src"
15
  if str(SRC_PATH) not in sys.path:
@@ -18,16 +21,31 @@ if str(SRC_PATH) not in sys.path:
18
  from twitter_sentiment.config import load_config
19
  from twitter_sentiment.predictor import load_artifacts, predict_with_threshold
20
 
21
- st.set_page_config(page_title="Twitter Sentiment Intelligence", page_icon="💼", layout="wide")
22
-
 
 
 
 
 
 
 
 
 
 
23
  @st.cache_resource(show_spinner=False)
24
  def _load_dependencies():
 
25
  config = load_config()
26
  pipeline, metrics = load_artifacts(config)
27
  return config, pipeline, metrics
28
 
29
 
 
 
 
30
  def format_probabilities(probabilities: Dict[str, float]) -> pd.DataFrame:
 
31
  return (
32
  pd.DataFrame.from_dict(probabilities, orient="index", columns=["confidence"])
33
  .sort_values("confidence", ascending=False)
@@ -35,54 +53,11 @@ def format_probabilities(probabilities: Dict[str, float]) -> pd.DataFrame:
35
  )
36
 
37
 
 
 
 
38
  def main() -> None:
 
39
  config, pipeline, metrics = _load_dependencies()
40
 
41
- st.title("Twitter Sentiment Intelligence Dashboard")
42
- st.caption(
43
- "Built to demonstrate Deloitte India Oracle Analyst competencies across data engineering, analytics, and automation."
44
- )
45
-
46
- with st.sidebar:
47
- st.header("Model Snapshot")
48
- st.write("**Classes:**", ", ".join(pipeline.classes_))
49
- if metrics:
50
- st.metric("Macro F1", f"{metrics.get('f1_macro', 0.0):.2f}")
51
- st.metric("Accuracy", f"{metrics.get('accuracy', 0.0):.2f}")
52
- st.download_button(
53
- "Download Metrics JSON",
54
- data=json.dumps(metrics or {}, indent=2).encode("utf-8"),
55
- file_name="metrics.json",
56
- mime="application/json",
57
- )
58
- st.info(
59
- "🚀 Tip: integrate with Oracle Autonomous Database by updating `config/settings.yaml`."
60
- )
61
-
62
- tab_predict, tab_metrics = st.tabs(["Predict", "Model Governance"])
63
-
64
- with tab_predict:
65
- st.subheader("Real-time Sentiment Assessment")
66
- user_input = st.text_area("Enter a tweet or customer verbatim", height=150)
67
- if st.button("Run Analysis", type="primary"):
68
- if not user_input.strip():
69
- st.warning("Please enter text to analyse.")
70
- else:
71
- label, probabilities = predict_with_threshold(user_input, config)
72
- st.success(f"Predicted sentiment: **{label.title()}**")
73
- st.dataframe(format_probabilities(probabilities), use_container_width=True)
74
-
75
- with tab_metrics:
76
- st.subheader("Operational Metrics")
77
- if metrics:
78
- metrics_df = pd.DataFrame(metrics, index=["score"]).T.rename(columns={"score": "value"})
79
- st.dataframe(metrics_df, use_container_width=True)
80
- else:
81
- st.info("Metrics will appear after the first training run (see `scripts/train.py`).")
82
-
83
- st.markdown("---")
84
- st.caption("© 2024 Deloitte-aligned Sentiment Analytics Accelerator")
85
-
86
-
87
- if __name__ == "__main__":
88
- main()
 
1
+ """Streamlit front-end for the Deloitte-ready Twitter Sentiment Intelligence dashboard."""
2
 
3
  from __future__ import annotations
4
 
5
  import json
 
6
  import sys
7
+ from pathlib import Path
8
  from typing import Dict
9
 
10
  import pandas as pd
11
  import streamlit as st
12
 
13
+ # -------------------------------------------------------------------------
14
+ # Path setup to include the local src/ package for imports
15
+ # -------------------------------------------------------------------------
16
  ROOT = Path(__file__).resolve().parents[1]
17
  SRC_PATH = ROOT / "src"
18
  if str(SRC_PATH) not in sys.path:
 
21
  from twitter_sentiment.config import load_config
22
  from twitter_sentiment.predictor import load_artifacts, predict_with_threshold
23
 
24
+ # -------------------------------------------------------------------------
25
+ # Streamlit App Configuration
26
+ # -------------------------------------------------------------------------
27
+ st.set_page_config(
28
+ page_title="Twitter Sentiment Intelligence",
29
+ page_icon="💼",
30
+ layout="wide",
31
+ )
32
+
33
+ # -------------------------------------------------------------------------
34
+ # Cached resource loading (config, pipeline, metrics)
35
+ # -------------------------------------------------------------------------
36
  @st.cache_resource(show_spinner=False)
37
  def _load_dependencies():
38
+ """Load configuration, trained pipeline, and metrics from artifacts."""
39
  config = load_config()
40
  pipeline, metrics = load_artifacts(config)
41
  return config, pipeline, metrics
42
 
43
 
44
+ # -------------------------------------------------------------------------
45
+ # Helper function to format prediction probabilities
46
+ # -------------------------------------------------------------------------
47
  def format_probabilities(probabilities: Dict[str, float]) -> pd.DataFrame:
48
+ """Convert prediction probabilities to a styled DataFrame for display."""
49
  return (
50
  pd.DataFrame.from_dict(probabilities, orient="index", columns=["confidence"])
51
  .sort_values("confidence", ascending=False)
 
53
  )
54
 
55
 
56
+ # -------------------------------------------------------------------------
57
+ # Main Streamlit Application
58
+ # -------------------------------------------------------------------------
59
  def main() -> None:
60
+ """Render the Deloitte-ready Twitter Sentiment Intelligence Dashboard."""
61
  config, pipeline, metrics = _load_dependencies()
62
 
63
+ # -----------------------------------------------------