Sensor Data Pattern Recognition Script
Budget: ₹1,500 – ₹12,500 INR
I have several Excel workbooks filled with time-stamped sensor readings and I need a compact AI-driven solution that can uncover recurring motifs and distinctive signatures hiding in that data. Your task is to create a Python-based script that ingests the .xlsx files, processes the time-series, and flags every instance of the patterns it detects—complete with confidence scores and clear start-end indices.
The raw files follow a simple structure: one column of timestamps followed by multiple numeric sensor channels recorded at regular intervals. Missing entries and small gaps are common, so the script must handle basic cleaning (forward/backward fill or interpolation). After that, feel free to choose the most effective approach—unsupervised clustering, shape-based matching, CNN, LSTM, or a hybrid—as long as the final output spotlights the repeating shapes and makes it easy for me to inspect them.
Key expectations
• Command-line execution with a single config or argument block (input path, channels to analyze, threshold tuning).
• Pattern summary report delivered as CSV or JSON, plus optional Matplotlib/Plotly visuals for quick validation.
. Annomolies detected
. Forecasting based on alert data for possible future alerts
• Clear, well-commented code using familiar libraries: pandas, NumPy, scikit-learn, TensorFlow or PyTorch (your call).
• Straightforward ways for me to tweak thresholds and extend to new sensors later.
Deliverables
- Python 3 script or module that runs on Windows without modification
- README explaining setup (virtualenv/conda), usage, and parameter descriptions
- Demonstration file or notebook showing the script correctly identifying at least three sample patterns I’ll provide
I will sign off once the script executes locally, reproduces the known patterns, and stays within PEP 8 conventions. Tell me which libraries you prefer and how quickly you can turn this around—I’m ready to get started.
The raw files follow a simple structure: one column of timestamps followed by multiple numeric sensor channels recorded at regular intervals. Missing entries and small gaps are common, so the script must handle basic cleaning (forward/backward fill or interpolation). After that, feel free to choose the most effective approach—unsupervised clustering, shape-based matching, CNN, LSTM, or a hybrid—as long as the final output spotlights the repeating shapes and makes it easy for me to inspect them.
Key expectations
• Command-line execution with a single config or argument block (input path, channels to analyze, threshold tuning).
• Pattern summary report delivered as CSV or JSON, plus optional Matplotlib/Plotly visuals for quick validation.
. Annomolies detected
. Forecasting based on alert data for possible future alerts
• Clear, well-commented code using familiar libraries: pandas, NumPy, scikit-learn, TensorFlow or PyTorch (your call).
• Straightforward ways for me to tweak thresholds and extend to new sensors later.
Deliverables
- Python 3 script or module that runs on Windows without modification
- README explaining setup (virtualenv/conda), usage, and parameter descriptions
- Demonstration file or notebook showing the script correctly identifying at least three sample patterns I’ll provide
I will sign off once the script executes locally, reproduces the known patterns, and stays within PEP 8 conventions. Tell me which libraries you prefer and how quickly you can turn this around—I’m ready to get started.