Update README.md

README.md

@@ -1,44 +1,84 @@
-license: openrail
-language:
-en
-metrics:
-accuracy
-pipeline_tag: image-classification
-base_model:
-tensorflow/keras
-training_process:
-CNN with 3 convolutional layers, max pooling, and 2 dense layers
-Trained on custom stock chart image dataset
-Binary classification (bullish vs bearish)
-datasets:
-Custom stock chart image dataset (train/val/test splits)
+stock_trend_analysis:
+  model:
+    architecture:
+      image_branch:
+        input_shape: [150, 150, 3]
+        layers:
+          - Conv2D_1:
+              filters: 32
+              kernel_size: [3, 3]
+              activation: relu
+          - MaxPooling2D_1:
+              pool_size: [2, 2]
+          - Conv2D_2:
+              filters: 64
+              kernel_size: [3, 3]
+              activation: relu
+          - MaxPooling2D_2:
+              pool_size: [2, 2]
+          - Conv2D_3:
+              filters: 128
+              kernel_size: [3, 3]
+              activation: relu
+          - MaxPooling2D_3:
+              pool_size: [2, 2]
+          - Flatten
 
-Model Overview
-This Convolutional Neural Network (CNN) classifies stock chart images as bullish or bearish, using a binary classification approach to predict market trends based on visual patterns.
+      numerical_branch:
+        input_shape: [5]  # 5 technical indicators
+        layers:
+          - Dense:
+              units: 64
+              activation: relu
 
-Model Details
-Architecture: Sequential CNN with 3 Convolutional layers, MaxPooling, Flatten, and 2 Dense layers
-Input: RGB images (150x150 pixels)
-Output: Binary classification (0 for bearish, 1 for bullish)
-Training Data: Custom dataset of stock chart images, split into train/validation/test sets
-Metrics: Accuracy, Precision, Recall, F1-Score
+      combined_layers:
+        - Dense:
+            units: 128
+            activation: relu
+        - Dense:
+            units: 1
+            activation: sigmoid
 
-Intended Use
-Primary Use: Automated classification of stock chart patterns
-Intended Users: Financial analysts, traders, automated trading systems
-Out-of-Scope Uses: Real-time trading decisions without human oversight, long-term market predictions
-Performance and Limitations
-Metrics: Accuracy: 0.57
+  data_generator:
+    class: CombinedDataGenerator
+    parameters:
+      batch_size: 32
+      img_size: [150, 150]
+      shuffle: true
+    
+    feature_extraction:
+      image_processing:
+        - grayscale_conversion
+        - edge_detection:
+            method: Canny
+            threshold1: 50
+            threshold2: 150
+        - line_detection:
+            method: HoughLinesP
+            min_line_length: 20
+            max_line_gap: 5
 
-Limitations:
-Binary classification only
-May not capture complex market dynamics
-Performance depends on training data quality
-Ethical Considerations
-Potential for dataset-based biases
-Not to be used as sole basis for financial decisions
-May perpetuate existing market trends
-Caveats and Recommendations
-Regular retraining with recent data recommended
-Use alongside other analysis tools and human expertise
-Consider expanding to multi-class classification
+      technical_indicators:
+        - slope:
+            method: polyfit
+            degree: 1
+        - volatility:
+            method: standard_deviation
+        - trend_strength:
+            method: mean_absolute_diff
+        - momentum:
+            method: price_difference
+            period: 10
+        - support_resistance:
+            method: min_max_difference
+
+  training:
+    optimizer: adam
+    loss: binary_crossentropy
+    metrics: [accuracy]
+
+  requirements:
+    - tensorflow
+    - opencv-python
+    - numpy
+    - scikit-learn