The paper presents DeepTrader, a deep learning neural network that automatically learns to become a high-performing algorithmic trader by passively observing an existing successful trader T. The system takes Level-2 limit order book (LOB) data as input and is trained on snapshot/quote pairs: for each LOB snapshot S observed by T at the time of issuing a quote Q, the network learns to output Q when shown S. This approach avoids explicit price prediction and instead directly mimics T's decision-making process. DeepTrader is evaluated against several benchmark algorithmic trading systems, including two that have been shown to outperform human traders, and it matches or exceeds their performance. Additionally, the authors analyze the trained network to identify which LOB features are most influential, providing interpretability. The work demonstrates a novel method for creating algorithmic traders via imitation learning from market data, with potential applications in automated trading strategy development.