Update README.md

README.md

@@ -1,3 +1,107 @@
 ---
 license: mit
+language:
+- en
+tags:
+- bittensor
 ---
+MIT License
+
+Copyright (c) 2024 Taoshi Inc
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+# Background
+
+The models provided here were created using open source modeling techniques
+provided in https://github.com/taoshidev/time-series-prediction-subnet (TSPS).
+They were achieved using the `runnable/miner_training.py`, and tested against
+existing models in `runnable/miner_testing.py`.
+
+> **Note**<br>
+This model requires the Feature Set Creator (FSC) functionality added in the
+latest release of the TSPS.
+
+# Build Strategy
+
+This section outlines the strategy used to build the models.
+
+## Understanding Dataset Used
+
+The dataset used to build the models can be generated using the
+`runnable/generate_historical_data.py`. A lookback period between June 2023 and
+January 2024 on the 5m interval was used to train the model. Recent data was
+used because it more closely correlates to the current market and
+macroeconomic conditions.
+
+Testing data was used between January 2024 and February 2024 to determine the
+performance of the models. This was tested using the `runnable/miner_testing.py`
+file with live historical data sources.
+
+
+## Understanding Model Creation
+
+As of now, the model only uses the following features to predict:
+- close
+- high
+- low
+- volume
+- time of day
+- time of week
+- time of month
+
+Other features from a wide range of feature sources are being added to TSPS
+infrastructure in the near future as improvements to the FSC.
+
+A variety of windows and parameters were tested and eliminated. The final 
+strategy to derive this model was the following:
+
+```
+model = BaseMiningModel(
+    filename="model_v5_1.h5",
+    mode="w",
+    feature_count=7,
+    sample_count=500,
+    prediction_feature_count=1,
+    prediction_count=10,
+    prediction_length=100,
+    layers=[
+        [1024, 0],
+        [1024, 0.3],
+    ],
+    learning_rate=0.000001,
+    dtype=Policy("mixed_float16"),
+)
+```
+
+The LSTM model has two stacked layers with a 0.3 dropout rate.
+
+## Understanding Training Decisions
+
+Training was done with 500 samples per scenario and 128 scenarios per batch,
+with 20 training epochs and 10 passes over the entire dataset. Additional
+epochs and passes were not found to improve the model's predictions.
+
+## Strategy to Predict
+
+The strategy to predict 100 closes of data into the future was to use 10
+predictions evenly spaced along the length of the prediction space, and then
+linearly interpolating between each prediction. By doing so, the model could
+learn to predict the general shape of the market movement, rather than
+predicting all 100.