Simple practice with one dimensional data 1

[HidekiInnan]

Define input data1 (n x 1) = {P(T-n+1), P(T-n+2), P(T-n+3), .., P(T)} answer: = 0 if P(T)-d <= P(T+m) <= P(T)+d (if the price after m minutes is within +- d pips) = 1 if P(T+m) > P(T)+d = 2 if P(T+m) < P(T)-d

example with USDJPY, 1 minute interval data, n=10, m=5, d=3 111.34 111.36 111.37 111.33 111.32 111.35 111.39 111.45 111.44 111.40 (current price) 1 (answer if the price after 5 minutes is more than 111.40 + 0.03)

[HidekiInnan]

Practice 1 Sugi: make 10000 random data from USDJPY, 1 minute interval, n=200, m=5, d=5 http://www.histdata.com/download-free-forex-data/?/metatrader/1-minute-bar-quotes perhaps, use data from 2010 to 2015 to learn, 2016 for testing the performance Metal: Which optimization algorithm would work best?

[rsugino]

I added revised data to ~/Dropbox/shared/ILDL/make_training_data. If requested, I will upload original data.

[heavywatal]

Which column should I use?

day        time  what?      what?      what?      what?      what?
2015.01.01 13:01 119.666000 119.666000 119.666000 119.666000 0
2015.01.01 13:05 119.692000 119.692000 119.692000 119.692000 0
2015.01.01 13:18 119.692000 119.692000 119.681000 119.681000 0

How should I interpolate missing rows? (e.g., 13:02, 13:03, and 13:04 have the same value as 13:01?)

[rsugino]

http://www.histdata.com/f-a-q/data-files-detailed-specification/

3rd column is open 4th column is high 5th column is low 6th column is close 6th - 3rd = dif

There is no change from 13:01 to 13:05 and from 13:06 to 13:18.

[HidekiInnan]

The close data (6th) are commonly used, so the input data should be an array of the 6th column. I don't see the answer data in Sugino's file yet. I would ask to revise the data with more columns. We don't need 6th - 3rd = dif for the 7th column. In the new data, The 7th column: The price AFTER 5min. The 8th column: The price AFTER 10min. The 9th column: The price AFTER 15min. The 10th column: The price AFTER 20min. The 11th column: The price AFTER 30min. The 12th column: The price AFTER 60min. These future data will be used for the answer.

[heavywatal]

train_neuralnet.py から読みやすい形でxとtをそれぞれ1000セット作ってみました。

DAT_MT_USDJPY_M1_2015.t.csv.gz DAT_MT_USDJPY_M1_2015.x.csv.gz

[HidekiInnan]

Great!