codelol / pstock2

pstock

Must use Python 3, mainly for its better exception printing.

Also depends on the following packages:

(1) yahoo finance api python package, for history prices https://pypi.python.org/pypi/yahoo-finance/1.1.4

(2) google finance api python package, for current prices because yahoo data has a 15-minute delay https://pypi.python.org/pypi/googlefinance

(3) tabulate for formatting https://pypi.python.org/pypi/tabulateuse

pstock2's Issues

parallelize data fetching

when fetching data for 1k+ symbols, it would take up to 30 minutes when it's done one by one. Should do it in parallel.

Recognize gaps

Gap ups and downs can be easily recognized by using current low/high and previous day's low/high.

Avoid refreshing when out of trading hours

Currently it keeps getting latest information every 60 seconds, but this is unnecessary while trading is not active.

Recognize volume breakout

Recognize 2 types of volume breakout
Type 1: current volume versus previous days' average volume. This is the most important;
Type 2: if market is still open, 'predict' the full day volume and alert if volume breaks out.

There should be a configurable threshold for how much can be a breakout, such as >20% of previous days' average volume.

Detect python version and reject Python 2.x

Since now the program must be executed by Python 3.x, we need 1 or 2 lines of code to detect python version, and quit gracefully if needed.

Network exception handling

when requesting quotes failed, program should retry

Need a config file

The first config to start with should be refresh frequency, which is currently hard coded to be 60 seconds. Program should read the config file periodically and apply new config on the fly.

Recognize single-day doji patterns

Some doji patterns are only determined by its current-day prices, these can be a start. Doji patterns that are related to multi-day prices can be the next step.

Here are some reference patterns

load csv file while getting its name

Currently csv files are touched in 2 steps:

1. get the names of all interested csv files about a symbol to figure out the last history we have. This is used to figure out the starting date to ask for new history info
2. load all files in memory to constructed an in-memory map, keyed by date
   Between these 2 separate steps, a request is made to download latest history file.

The 2 steps should be combined into one giant step:

1. list files that are interesting, and for each file, immediately add its content to in-memory map.
2. get the entry with the biggest key (or, the largest date value). This is the latest history we have
   The second step will solve a bug that file name doesn't faithfully reflect the latest history we have.

time zone

History date is based on host's local time, but should be based on the local time of the exchange, such as US/Eastern.