Update README.md
Browse files
README.md
CHANGED
@@ -17,6 +17,7 @@ size_categories:
|
|
17 |
# Epsilon-Transformers Belief Analysis Dataset
|
18 |
|
19 |
This dataset contains trained neural network models and their corresponding belief state regression analysis from the Epsilon-Transformers project. The models were trained on four different stochastic processes and analyzed for their ability to learn and represent belief states.
|
|
|
20 |
|
21 |
## Dataset Structure
|
22 |
|
@@ -61,92 +62,13 @@ epsilon-transformers-belief-analysis/
|
|
61 |
| 20250421221507 | 0 | Transformer | Moon Process | Transformer trained on Moon Process |
|
62 |
| 20250422023003 | 1 | Transformer | FRDN | Transformer trained on FRDN |
|
63 |
|
64 |
-
## Process Descriptions
|
65 |
-
|
66 |
-
### Mess3 (Classical Process)
|
67 |
-
A classical stochastic process used as a baseline for comparison with quantum processes.
|
68 |
-
|
69 |
-
### FRDN (Finite Random Dynamics Networks)
|
70 |
-
A quantum process representing finite random dynamics networks, modeling quantum systems with specific structural properties.
|
71 |
-
|
72 |
-
### Bloch Walk
|
73 |
-
A quantum random walk process on the Bloch sphere, representing quantum state evolution in a geometric framework.
|
74 |
-
|
75 |
-
### Moon Process
|
76 |
-
A post-quantum stochastic process that explores computational mechanics beyond standard quantum frameworks.
|
77 |
-
|
78 |
-
## Model Architectures
|
79 |
-
|
80 |
-
### RNN Models (LSTM, GRU, RNN)
|
81 |
-
- **Layers**: 4
|
82 |
-
- **Hidden Units**: 64
|
83 |
-
- **Direction**: Unidirectional
|
84 |
-
- **Configuration**: L4_H64_uni
|
85 |
-
|
86 |
-
### Transformer Models
|
87 |
-
- **Layers**: 4
|
88 |
-
- **Attention Heads**: 4
|
89 |
-
- **Head Dimension**: 16
|
90 |
-
- **Model Dimension**: 64
|
91 |
-
- **Configuration**: L4_H4_DH16_DM64
|
92 |
-
|
93 |
## File Formats
|
94 |
|
95 |
### Model Files (.pt)
|
96 |
-
|
97 |
|
98 |
### Analysis Files (.joblib)
|
99 |
Joblib-serialized files containing:
|
100 |
- **checkpoint_*.joblib**: Regression analysis results mapping activations to belief states
|
101 |
- **ground_truth_data.joblib**: True belief states and probabilities for the neural network data
|
102 |
-
- **markov3_*.joblib**: Classical Markov model comparisons and baselines
|
103 |
-
|
104 |
-
## Usage
|
105 |
-
|
106 |
-
### Loading Models
|
107 |
-
|
108 |
-
```python
|
109 |
-
import torch
|
110 |
-
from pathlib import Path
|
111 |
-
|
112 |
-
# Load a model checkpoint
|
113 |
-
model_path = Path("models/20241121152808_57/4075724800.pt")
|
114 |
-
checkpoint = torch.load(model_path, map_location='cpu')
|
115 |
-
```
|
116 |
-
|
117 |
-
### Loading Analysis Data
|
118 |
-
|
119 |
-
```python
|
120 |
-
import joblib
|
121 |
-
from pathlib import Path
|
122 |
-
|
123 |
-
# Load regression analysis results
|
124 |
-
analysis_path = Path("analysis/20241121152808_57/checkpoint_4075724800.joblib")
|
125 |
-
analysis_data = joblib.load(analysis_path)
|
126 |
-
|
127 |
-
# Access layer-wise regression metrics
|
128 |
-
for layer, metrics in analysis_data.items():
|
129 |
-
print(f"Layer {layer} RMSE: {metrics['rmse']}")
|
130 |
-
```
|
131 |
-
|
132 |
-
## Citation
|
133 |
-
|
134 |
-
If you use this dataset in your research, please cite:
|
135 |
-
|
136 |
-
```bibtex
|
137 |
-
@misc{epsilon-transformers-belief-analysis,
|
138 |
-
title={Epsilon-Transformers Belief Analysis Dataset},
|
139 |
-
author={[Your Name]},
|
140 |
-
year={2024},
|
141 |
-
howpublished={Hugging Face Datasets},
|
142 |
-
url={https://huggingface.co/datasets/[your-username]/epsilon-transformers-belief-analysis}
|
143 |
-
}
|
144 |
-
```
|
145 |
-
|
146 |
-
## License
|
147 |
-
|
148 |
-
[Specify your license here]
|
149 |
-
|
150 |
-
## Contact
|
151 |
-
|
152 |
-
[Your contact information]
|
|
|
17 |
# Epsilon-Transformers Belief Analysis Dataset
|
18 |
|
19 |
This dataset contains trained neural network models and their corresponding belief state regression analysis from the Epsilon-Transformers project. The models were trained on four different stochastic processes and analyzed for their ability to learn and represent belief states.
|
20 |
+
See https://github.com/adamimos/epsilon-transformers/tree/quantum-public for codebase which generated this data.
|
21 |
|
22 |
## Dataset Structure
|
23 |
|
|
|
62 |
| 20250421221507 | 0 | Transformer | Moon Process | Transformer trained on Moon Process |
|
63 |
| 20250422023003 | 1 | Transformer | FRDN | Transformer trained on FRDN |
|
64 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
## File Formats
|
66 |
|
67 |
### Model Files (.pt)
|
68 |
+
Transformerlens (for transformers) or Pytorch (for RNNs) model checkpoints containing trained model weights and optimizer states.
|
69 |
|
70 |
### Analysis Files (.joblib)
|
71 |
Joblib-serialized files containing:
|
72 |
- **checkpoint_*.joblib**: Regression analysis results mapping activations to belief states
|
73 |
- **ground_truth_data.joblib**: True belief states and probabilities for the neural network data
|
74 |
+
- **markov3_*.joblib**: Classical Markov model comparisons and baselines
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|