1. If you still need something in Deep Learning, you can consider Theano or Keras.

These two things may be used in analyzing news data.

However, it is not recommended to use this method to build quantitative models, because the calculation amount is too large and the cost is very high.

2. In terms of trading framework, in addition to vn.py, we also recommend the PyAlgoTrade framework, which can be found on github.

I personally think that this framework is much more powerful than vn.py. After all, it is the work of an old man who has been in the financial IT field for nearly 20 years. The architectural design is not generally excellent.

3. Domestically, Ricequant is a good choice. Although they use Java, I have seen the team. They are all from financial IT backgrounds. Basically, they have more than 7 or 8 years of experience. They have very solid basic skills and can do everything well.

Very reliable.

Now they are also considering extending the SDK to Python.

4. The domestic market and trading interfaces use their own protocols (for example, the CTP interface uses the FTD protocol) instead of the FIX protocol that is widely used internationally, and they are not open source.

If you need to connect to the market, you also need to consider encapsulating the interface SDK in Python.

(Edit: Someone mentioned in the comments that many brokerages have also opened the FIX interface, but it seems to be used on the intranet) 5. Someone talked about the database. Here I will also say that for high-frequency tick-level data, its magnitude can be

Reaching the TB level per day, ordinary relational databases cannot handle it.

If you try to use a traditional relational database, such as Oracle, you can save it.

For this level of data, it would be better to use file system + memory index.

However, this kind of scenario is generally encountered in institutions, and individual quants do not need to consider it.