Add ALIGNN work to separate branch

Author: sidnb13

matdeeplearn/common/metrics.py

@@ -0,0 +1,84 @@
+import matplotlib.pyplot as plt
+import pandas as pd
+import os
+from datetime import datetime
+
+root_path = '/global/cfs/projectdirs/m3641/Sidharth/MatDeepLearn_dev/output/'
+
+class MetricMonitor:
+    '''
+    Monitor and update various training metrics.
+    '''
+    def __init__(self, plotpath=os.path.join(root_path, 'plots'), datapath=os.path.join(root_path, 'data'), epoch_step=5, start_epoch=0) -> None:
+        self.streams = {}
+        self.epoch = {}
+        self.start_epoch = start_epoch
+        self.epoch_step = epoch_step
+        self.plotpath = plotpath
+        self.datapath = datapath
+    
+    def create_data_stream(self, stream_name):
+        '''
+        Create a metric data stream to be updated.
+        '''
+        self.streams[stream_name] = []
+        self.epoch[stream_name] = [self.start_epoch]
+    
+    def update(self, stream_name, val):
+        '''
+        Update a data stream with a specified value.
+        '''
+        if stream_name in self.streams:
+            self.streams[stream_name].append(val)             
+            self.epoch[stream_name].append(self.epoch[stream_name][len(self.epoch[stream_name]) - 1] + self.epoch_step)
+    
+    def save_outputs(self):
+        '''
+        Save post-training metric outputs.
+        '''
+        timestamp = datetime.now()
+            
+        metric_df = pd.DataFrame(self.streams)
+        print(metric_df)
+        print(os.path.join(self.datapath, f'train_metric_{timestamp}.csv'))
+        metric_df.to_csv(os.path.join(self.datapath, f'train_metric_{timestamp}.csv'))
+        
+        fig, axs = plt.subplots(len(self.streams.keys()))
+        fig.suptitle(f'Training metrics {timestamp}')
+        
+        for ax, item in zip(axs, self.streams.items()):
+            ax.plot(self.epoch[item[0]][:-1], item[1])
+            ax.set_title(item[0])
+            
+        print(os.path.join(self.plotpath, f'plot_metrics_{timestamp}.png'))
+        plt.savefig(os.path.join(self.plotpath, f'plot_metrics_{timestamp}.png'))
+
+class DatasetMetrics:
+    '''
+    Analyze a graph dataset for basic properties
+    and create basic visualization of overall statistics.
+    '''
+    def __init__(self) -> None:
+        pass
+
+class VisualizeGraph:
+    '''
+    Visualize input and latent space graphs with heatmap plots.
+    TODO: Look at old MatDeepLearn to port over latent visualization code.
+    ''' 
+    def __init__(self) -> None:
+        pass
+
+# Testing code
+if __name__ == '__main__':    
+    print('Testing metrics')
+    
+    m = MetricMonitor()
+    m.create_data_stream('test1')
+    m.create_data_stream('test2')
+    
+    for i in range(100):
+        m.update('test1', i)
+        m.update('test2', i)
+        
+    m.save_outputs()