Updated README.md

README.md

@@ -11,22 +11,18 @@ The CNN is built with state-of-the-art architectures, such as [EfficientNet (Tan
 
 
 ## Prerequisites
-- Python 3 (tested on Python 3.6 and 3.7)
-- TensorFlow (1.14 or above, 2.1 or above)
-- Scikit-learn
-- Scikit-image
-- OpenCV-Python
-- h5py
+- Python 3 (tested on Python 3.8)
+- TensorFlow 2.x (tested on TensorFlow 2.3.1)
+- Scikit-learn (tested on 0.24.1)
+- Scikit-image (tested on 0.18.1)
+- OpenCV-Python (tested on 4.5.1.48)
+- h5py (tested on 2.10.0)
+- keras-applications (tested on 1.0.8)
 - Horovod (optional; 0.19 or above)
 - MPI (OpenMPI/MPICH, required when Horovod is installed)
 
+It is recommended to install the prerequisites within a virtual environment.
 
-Use the package manager conda to install dependencies.
-
-```
-conda install -r requirements.txt -n tf2gpu
-conda activate tf2gpu
-```
 
 ## Training data
 We simulated galaxy mergers of different mass ratios and size ratios (1:1, 1:2, 1:3, 2:3). The snapshots of the simulations are visualized once every 5 Myr (million years) using cameras from 14 different positions, and therefore generating 14 2D images. Each simulation should complete within a timescale of about 700 Myr. These images are stored in a compressed HDF5 dataset. The available image resolution are (256, 256), (512, 512), (1024, 1024), and (2048, 2048) pixels. The dataset is balanced.
@@ -71,6 +67,25 @@ When high-resolution images are trained on a large DNN, the memory consumption o
 
 Please note that this option usually comes with performance penalty. Alternatively, running large models on CPUs may actually be faster in certain hardware configurations.
 
+## Use DeepGalaxy as a benchmark suite
+DeepGalaxy provides benchmark information (throughput) for the underlying hardware system. In the `train_log.txt` output file, the throughput of the code looks like this
+```
+[Performance] Epoch 0 takes 107.62 seconds. Throughput: 2.37 images/sec (per node), 9.48 images/sec (total)
+[Performance] Epoch 1 takes 17.07 seconds. Throughput: 14.94 images/sec (per node), 59.75 images/sec (total)
+[Performance] Epoch 2 takes 10.97 seconds. Throughput: 23.24 images/sec (per node), 92.94 images/sec (total)
+[Performance] Epoch 3 takes 11.01 seconds. Throughput: 23.16 images/sec (per node), 92.63 images/sec (total)
+[Performance] Epoch 4 takes 11.02 seconds. Throughput: 23.14 images/sec (per node), 92.56 images/sec (total)
+[Performance] Epoch 5 takes 10.82 seconds. Throughput: 23.57 images/sec (per node), 94.27 images/sec (total)
+[Performance] Epoch 6 takes 10.86 seconds. Throughput: 23.48 images/sec (per node), 93.92 images/sec (total)
+[Performance] Epoch 7 takes 10.95 seconds. Throughput: 23.29 images/sec (per node), 93.17 images/sec (total)
+[Performance] Epoch 8 takes 10.93 seconds. Throughput: 23.33 images/sec (per node), 93.32 images/sec (total)
+[Performance] Epoch 9 takes 11.01 seconds. Throughput: 23.17 images/sec (per node), 92.67 images/sec (total)
+```
+The above performance log gives insights into the throughput per node and the total throughput (if trained with multiple nodes/processors). Typically, the first 2-3 epochs have lower throughput due to the initialization effect. As such, the throughput should be read after the 3rd epoch when the throughput becomes stable. 
+
+By varying the number of workers (`-np` arguments, see above) and plot the corresponding total throughput as a function of `-np`, one can obtain a figure of scaling efficiency. Ideally, the total throughput scales linearly as a function of `-np`.  Practically, when `-np` is low, the scaling behavior is nearly linear, but the overhead picks up for large `-np` due to the communication costs in the `Allreduce` data parallel training. 
+
+
 ## Acknowledgement
 This project is supported by [PRACE](https://prace-ri.eu/), [SURF](https://www.surf.nl/en), [Intel PCC](https://software.intel.com/content/www/us/en/develop/topics/parallel-computing-centers.html), and [Leiden Observatory](https://www.universiteitleiden.nl/en/science/astronomy).