File size: 4,434 Bytes
3961ee7 e204e43 3961ee7 e204e43 3961ee7 3f91261 |
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 |
---
license: mit
language: en
tags:
- materials science
- synthesis prediction
- lightgbm
- cheminformatics
datasets: []
metrics:
- accuracy
- f1
---
# Synthesis Condition Predictor
This model predicts optimal temperature bins and atmosphere categories for inorganic material synthesis.
It was trained on a dataset of text-mined synthesis procedures. Here is the source of the dataset: https://www.nature.com/articles/s41597-019-0224-1
**Models Included:**
* Temperature Bin Prediction (LightGBM)
* Atmosphere Category Prediction (LightGBM)
**Intended Use:**
To assist researchers in designing synthesis experiments by predicting key process parameters.
Input a target material, precursors, and basic operational details to get predictions.
**How to Use:**
```python
# Ensure your inference script and its dependencies are in the PYTHONPATH
# from synthesis_predictor_hf_repo.src.inference import predict_synthesis_outcome, load_all_artifacts_once
# Or, if running from a cloned repo where 'src' is a subdirectory:
# from src.inference import predict_synthesis_outcome, load_all_artifacts_once
# if not load_all_artifacts_once():
# print("Failed to load model artifacts.")
# else:
# raw_input_example = {
# 'target_formula_raw': "YBa2Cu3O7",
# 'precursor_formulas_raw': ["Y2O3", "BaCO3", "CuO"],
# 'operations_simplified_list': [
# {'type': 'MixingOperation', 'string': 'Ball milling for 2h', 'conditions': {'duration': [{'value':2, 'unit':'h'}]}},
# {'type': 'HeatingOperation', 'string': 'Calcined at 920C for 10h in air',
# 'conditions': {'heating_temperature': [{'value':920}], 'heating_time': [{'value':10}], 'atmosphere':'air'}},
# {'type': 'HeatingOperation', 'string': 'Sintered at 950C for 20h in O2',
# 'conditions': {'heating_temperature': [{'value':950}], 'heating_time': [{'value':20}], 'atmosphere':'Oxygen'}}
# ],
# 'reactants_coeffs': [("Y2O3", 0.5), ("BaCO3", 2.0), ("CuO", 3.0)], # Example, adjust as needed
# 'products_coeffs': [("YBa2Cu3O7", 1.0)] # Example
# }
# predictions = predict_synthesis_outcome(raw_input_example)
# print(predictions)
```
**Limitations:**
* The model's accuracy is around 68-72%.
* Predictions are based on patterns in the training data and may not generalize to all chemical systems.
* The feature engineering for process parameters in the inference script relies on the user providing an `operations_simplified_list` that can be parsed by the internal logic. The quality of these inputs directly affects prediction accuracy.
**Training Data:**
The model was trained on a proprietary dataset of text-mined inorganic synthesis procedures. (Kononova et al.)
https://www.nature.com/articles/s41597-019-0224-1
**Evaluation Results:**
The models were evaluated on a hold-out test set.
**1. Tuned Temperature Bin Prediction Model:**
* **Overall Test Set Accuracy:** 0.6821
* **Overall Test Set F1 Score (Weighted):** 0.6785
* **Per-Class Performance (Test Set):**
```
precision recall f1-score support
TempBin_1_(1_to_900] 0.77 0.79 0.78 954
TempBin_2_(900_to_1100] 0.62 0.53 0.57 743
TempBin_3_(1100_to_1300] 0.58 0.58 0.58 768
TempBin_4_(1300_to_3000] 0.72 0.80 0.76 715
accuracy 0.68 3180
macro avg 0.67 0.68 0.67 3180
weighted avg 0.68 0.68 0.68 3180
```
**2. Tuned Atmosphere Category Prediction Model:**
* **Overall Test Set Accuracy:** 0.7193
* **Overall Test Set F1 Score (Weighted):** 0.7174
* **Per-Class Performance (Test Set):**
```
precision recall f1-score support
Inert 0.59 0.38 0.46 139
Other_Atm_Target 1.00 0.44 0.62 9
Oxidizing 0.67 0.71 0.69 1552
Reducing 0.70 0.47 0.56 100
Unknown_Atm_Category 0.76 0.76 0.76 2098
accuracy 0.72 3898
macro avg 0.74 0.55 0.62 3898
weighted avg 0.72 0.72 0.72 3898
```
)
|