Predicting stock prices can be a challenging task as it often does not follow any specific pattern. However, deep neural learning can be used to identify patterns through machine learning. One of the most effective techniques for series forecasting is using LSTM (long short-term memory) networks, which are a type of recurrent neural network (RNN) capable of remembering information over a long period of time. This makes them extremely useful for predicting stock prices.
This TensorFlow implementation of an LSTM neural network can be used for time series forecasting. Successful prediction of a stock's future price can yield significant profits for investors.
Predicting stock prices is a complex task, as it is influenced by various factors such as market trends, political events, and economic indicators. The fluctuations in the stock prices are driven by the forces of supply and demand, which can be unpredictable at times. To identify patterns and trends in stock prices, deep learning techniques can be used for machine learning. Long short-term memory (LSTM) is a type of recurrent neural network (RNN) that is specifically designed for sequence modeling and prediction. LSTM is capable of retaining information over an extended period of time, making it an ideal approach for predicting stock prices. As a result, RNNs are well-suited to time series data, where they process data step-by-step, maintaining an internal state where they store the information they have seen so far in a compressed form. Accurate prediction of a stock's future price can provide significant financial gain to investors.
To gather the necessary market data for our stock prediction model, we will utilize the yFinance library in Python. This library is designed specifically for downloading relevant information on a given ticker symbol from the Yahoo Finance Finance webpage. By using yFinance, we can easily access the latest market data and incorporate it into our model.
For our purposes, we will be using the ticker symbol "GOOG", which is a well-known technology company. Here's an example screenshot of the ticker symbol on the Yahoo Finance page:
To download the data info, we will need the yFinance
library installed and then we will only need to perform the following operation to download all the relevant information of a given Stock using its ticker symbol. Please make sure you use the latest version of the library (pip install yfinance==0.2.33
) as I have seen errors with previous versions.
Below is the output from the [download_market_data_info.py] file that is able to download financial data from Yahoo Finance.
C:\Users\thund\Source\Repos\stock-prediction-deep-neural-learning>python download_market_data_info.py
Info
{
"52WeekChange": 0.26037383,
"SandP52WeekChange": 0.034871936,
"address1": "1600 Amphitheatre Parkway",
"algorithm": null,
"annualHoldingsTurnover": null,
"annualReportExpenseRatio": null,
"ask": 1432.77,
"askSize": 1400,
"averageDailyVolume10Day": 2011171,
"averageVolume": 1857809,
"averageVolume10days": 2011171,
"beta": 1.068946,
"beta3Year": null,
"bid": 1432.16,
"bidSize": 3000,
"bookValue": 297.759,
"category": null,
"circulatingSupply": null,
"city": "Mountain View",
"companyOfficers": [],
"country": "United States",
"currency": "USD",
"dateShortInterest": 1592179200,
"dayHigh": 1441.19,
"dayLow": 1409.82,
"dividendRate": null,
"dividendYield": null,
"earningsQuarterlyGrowth": 0.027,
"enterpriseToEbitda": 17.899,
"enterpriseToRevenue": 5.187,
"enterpriseValue": 864533741568,
"exDividendDate": null,
"exchange": "NMS",
"exchangeTimezoneName": "America/New_York",
"exchangeTimezoneShortName": "EDT",
"expireDate": null,
"fiftyDayAverage": 1417.009,
"fiftyTwoWeekHigh": 1532.106,
"fiftyTwoWeekLow": 1013.536,
"fiveYearAverageReturn": null,
"fiveYearAvgDividendYield": null,
"floatShares": 613293304,
"forwardEps": 55.05,
"forwardPE": 26.028149,
"fromCurrency": null,
"fullTimeEmployees": 123048,
"fundFamily": null,
"fundInceptionDate": null,
"gmtOffSetMilliseconds": "-14400000",
"heldPercentInsiders": 0.05746,
"heldPercentInstitutions": 0.7062,
"industry": "Internet Content & Information",
"isEsgPopulated": false,
"lastCapGain": null,
"lastDividendValue": null,
"lastFiscalYearEnd": 1577750400,
"lastMarket": null,
"lastSplitDate": 1430092800,
"lastSplitFactor": "10000000:10000000",
"legalType": null,
"logo_url": "https://logo.clearbit.com/abc.xyz",
"longBusinessSummary": "Alphabet Inc. provides online advertising services in the United States, Europe, the Middle East, Africa, the Asia-Pacific, Canada, and Latin America. It offers performance and brand advertising services. The company operates through Google and Other Bets segments. The Google segment offers products, such as Ads, Android, Chrome, Google Cloud, Google Maps, Google Play, Hardware, Search, and YouTube, as well as technical infrastructure. It also offers digital content, cloud services, hardware devices, and other miscellaneous products and services. The Other Bets segment includes businesses, including Access, Calico, CapitalG, GV, Verily, Waymo, and X, as well as Internet and television services. Alphabet Inc. was founded in 1998 and is headquartered in Mountain View, California.",
"longName": "Alphabet Inc.",
"market": "us_market",
"marketCap": 979650805760,
"maxAge": 1,
"maxSupply": null,
"messageBoardId": "finmb_29096",
"morningStarOverallRating": null,
"morningStarRiskRating": null,
"mostRecentQuarter": 1585612800,
"navPrice": null,
"netIncomeToCommon": 34522001408,
"nextFiscalYearEnd": 1640908800,
"open": 1411.1,
"openInterest": null,
"payoutRatio": 0,
"pegRatio": 4.38,
"phone": "650-253-0000",
"previousClose": 1413.61,
"priceHint": 2,
"priceToBook": 4.812112,
"priceToSalesTrailing12Months": 5.87754,
"profitMargins": 0.20712,
"quoteType": "EQUITY",
"regularMarketDayHigh": 1441.19,
"regularMarketDayLow": 1409.82,
"regularMarketOpen": 1411.1,
"regularMarketPreviousClose": 1413.61,
"regularMarketPrice": 1411.1,
"regularMarketVolume": 1084440,
"revenueQuarterlyGrowth": null,
"sector": "Communication Services",
"sharesOutstanding": 336161984,
"sharesPercentSharesOut": 0.0049,
"sharesShort": 3371476,
"sharesShortPreviousMonthDate": 1589500800,
"sharesShortPriorMonth": 3462105,
"shortName": "Alphabet Inc.",
"shortPercentOfFloat": null,
"shortRatio": 1.9,
"startDate": null,
"state": "CA",
"strikePrice": null,
"symbol": "GOOG",
"threeYearAverageReturn": null,
"toCurrency": null,
"totalAssets": null,
"tradeable": false,
"trailingAnnualDividendRate": null,
"trailingAnnualDividendYield": null,
"trailingEps": 49.572,
"trailingPE": 28.904415,
"twoHundredDayAverage": 1352.9939,
"volume": 1084440,
"volume24Hr": null,
"volumeAllCurrencies": null,
"website": "https://www.abc.xyz",
"yield": null,
"ytdReturn": null,
"zip": "94043"
}
ISIN
-
Major Holders
0 1
0 5.75% % of Shares Held by All Insider
1 70.62% % of Shares Held by Institutions
2 74.93% % of Float Held by Institutions
3 3304 Number of Institutions Holding Shares
Institutional Holders
Holder Shares Date Reported % Out Value
0 Vanguard Group, Inc. (The) 23162950 2020-03-30 0.0687 26934109889
1 Blackrock Inc. 20264225 2020-03-30 0.0601 23563443472
2 Price (T.Rowe) Associates Inc 12520058 2020-03-30 0.0371 14558448642
3 State Street Corporation 11814026 2020-03-30 0.0350 13737467573
4 FMR, LLC 8331868 2020-03-30 0.0247 9688379429
5 Capital International Investors 4555880 2020-03-30 0.0135 5297622822
6 Geode Capital Management, LLC 4403934 2020-03-30 0.0131 5120938494
7 Northern Trust Corporation 4017009 2020-03-30 0.0119 4671018235
8 JP Morgan Chase & Company 3707376 2020-03-30 0.0110 4310973886
9 AllianceBernstein, L.P. 3483382 2020-03-30 0.0103 4050511423
Dividents
Series([], Name: Dividends, dtype: int64)
Splits
Date
2014-03-27 2.002
2015-04-27 1.000
Name: Stock Splits, dtype: float64
Actions
Dividends Stock Splits
Date
2014-03-27 0.0 2.002
2015-04-27 0.0 1.000
Calendar
Empty DataFrame
Columns: []
Index: [Earnings Date, Earnings Average, Earnings Low, Earnings High, Revenue Average, Revenue Low, Revenue High]
Recommendations
Firm To Grade From Grade Action
Date
2012-03-14 15:28:00 Oxen Group Hold init
2012-03-28 06:29:00 Citigroup Buy main
2012-04-03 08:45:00 Global Equities Research Overweight main
2012-04-05 06:34:00 Deutsche Bank Buy main
2012-04-09 06:03:00 Pivotal Research Buy main
2012-04-10 11:32:00 UBS Buy main
2012-04-13 06:16:00 Deutsche Bank Buy main
2012-04-13 06:18:00 Jefferies Buy main
2012-04-13 06:37:00 PiperJaffray Overweight main
2012-04-13 06:38:00 Goldman Sachs Neutral main
2012-04-13 06:41:00 JP Morgan Overweight main
2012-04-13 06:51:00 Oppenheimer Outperform main
2012-04-13 07:13:00 Benchmark Hold main
2012-04-13 08:46:00 BMO Capital Outperform main
2012-04-16 06:52:00 Hilliard Lyons Buy main
2012-06-06 06:17:00 Deutsche Bank Buy main
2012-06-06 06:56:00 JP Morgan Overweight main
2012-06-22 06:15:00 Citigroup Buy main
2012-07-13 05:57:00 Wedbush Neutral init
2012-07-17 09:33:00 Outperform main
2012-07-20 06:43:00 Benchmark Hold main
2012-07-20 06:54:00 Deutsche Bank Buy main
2012-07-20 06:59:00 Bank of America Buy main
2012-08-13 05:49:00 Morgan Stanley Overweight Equal-Weight up
2012-09-17 06:07:00 Global Equities Research Overweight main
2012-09-21 06:28:00 Cantor Fitzgerald Buy init
2012-09-24 06:11:00 Citigroup Buy main
2012-09-24 09:05:00 Pivotal Research Buy main
2012-09-25 07:20:00 Capstone Buy main
2012-09-26 05:48:00 Canaccord Genuity Buy main
... ... ... ... ...
2017-10-27 19:29:31 UBS Buy main
2018-02-02 14:04:52 PiperJaffray Overweight Overweight main
2018-04-24 11:43:49 JP Morgan Overweight Overweight main
2018-04-24 12:24:37 Deutsche Bank Buy Buy main
2018-05-05 14:00:37 B. Riley FBR Buy main
2018-07-13 13:49:13 Cowen & Co. Outperform Outperform main
2018-07-24 11:50:55 Cowen & Co. Outperform Outperform main
2018-07-24 13:33:47 Raymond James Outperform Outperform main
2018-10-23 11:18:00 Deutsche Bank Buy Buy main
2018-10-26 15:17:08 Raymond James Outperform Outperform main
2019-01-23 12:55:04 Deutsche Bank Buy Buy main
2019-02-05 12:55:12 Deutsche Bank Buy Buy main
2019-02-05 13:18:47 PiperJaffray Overweight Overweight main
2019-05-15 12:34:54 Deutsche Bank Buy main
2019-10-23 12:58:59 Credit Suisse Outperform main
2019-10-29 11:58:09 Raymond James Outperform main
2019-10-29 14:15:40 Deutsche Bank Buy main
2019-10-29 15:48:29 UBS Buy main
2020-01-06 11:22:07 Pivotal Research Buy Hold up
2020-01-17 13:01:48 UBS Buy main
2020-02-04 12:26:56 Piper Sandler Overweight main
2020-02-04 12:41:00 Raymond James Outperform main
2020-02-04 14:00:36 Deutsche Bank Buy main
2020-02-06 11:34:20 CFRA Strong Buy main
2020-03-18 13:52:51 JP Morgan Overweight main
2020-03-30 13:26:16 UBS Buy main
2020-04-17 13:01:41 Oppenheimer Outperform main
2020-04-20 19:29:50 Credit Suisse Outperform main
2020-04-29 14:01:51 UBS Buy main
2020-05-05 12:44:16 Deutsche Bank Buy main
[219 rows x 4 columns]
Earnings
Empty DataFrame
Columns: [Open, High, Low, Close, Adj Close, Volume]
Index: []
Quarterly Earnings
Empty DataFrame
Columns: [Open, High, Low, Close, Adj Close, Volume]
Index: []
Financials
Empty DataFrame
Columns: [Open, High, Low, Close, Adj Close, Volume]
Index: []
Quarterly Financials
Empty DataFrame
Columns: [Open, High, Low, Close, Adj Close, Volume]
Index: []
Balance Sheet
Empty DataFrame
Columns: [Open, High, Low, Close, Adj Close, Volume]
Index: []
Quarterly Balance Sheet
Empty DataFrame
Columns: [Open, High, Low, Close, Adj Close, Volume]
Index: []
Balancesheet
Empty DataFrame
Columns: [Open, High, Low, Close, Adj Close, Volume]
Index: []
Quarterly Balancesheet
Empty DataFrame
Columns: [Open, High, Low, Close, Adj Close, Volume]
Index: []
Cashflow
Empty DataFrame
Columns: [Open, High, Low, Close, Adj Close, Volume]
Index: []
Quarterly Cashflow
Empty DataFrame
Columns: [Open, High, Low, Close, Adj Close, Volume]
Index: []
Sustainability
None
Options
('2020-07-02', '2020-07-10', '2020-07-17', '2020-07-24', '2020-07-31', '2020-08-07', '2020-08-21', '2020-09-18', '2020-11-20', '2020-12-01', '2020-12-18', '2021-01-15', '2021-06-18', '2022-01-21', '2022-06-17')
The data has a JSON document that we could use later on to create our Security Master if we ever wanted to store this data somewhere to keep track of the Securities we are going to trade with. As the data could come with different fields, my suggestion is to store them on a Data Lake so we can build it from multiple sources without having to worry too much about the way the data is structured.
The previous step helps us to identify several characteristics of a given ticker symbol so we can use its properties to define some of the charts I'm showing below. Note that the yFinance library only requires the stock to download via ticker symbol, the start date and end date of the period we want to get. Additionally, we can also specify the granularity of the data using the interval parameter. By default, the interval is 1 day and this is the one I will use for my training.
To download the data we can use the following command:
start = pd.to_datetime('2004-08-01')
stock = ['GOOG']
data = yf.download(stock, start=start, end=datetime.date.today())
print(data)
And the sample output:
C:\Users\thund\Source\Repos\stock-prediction-deep-neural-learning>python download_market_data.py
[*********************100%***********************] 1 of 1 completed
Open High Low Close Adj Close Volume
Date
2004-08-19 49.813286 51.835709 47.800831 49.982655 49.982655 44871300
2004-08-20 50.316402 54.336334 50.062355 53.952770 53.952770 22942800
2004-08-23 55.168217 56.528118 54.321388 54.495735 54.495735 18342800
2004-08-24 55.412300 55.591629 51.591621 52.239193 52.239193 15319700
2004-08-25 52.284027 53.798351 51.746044 52.802086 52.802086 9232100
2004-08-26 52.279045 53.773445 52.134586 53.753517 53.753517 7128600
2004-08-27 53.848164 54.107193 52.647663 52.876804 52.876804 6241200
2004-08-30 52.443428 52.548038 50.814533 50.814533 50.814533 5221400
2004-08-31 50.958992 51.661362 50.889256 50.993862 50.993862 4941200
2004-09-01 51.158245 51.292744 49.648903 49.937820 49.937820 9181600
2004-09-02 49.409801 50.993862 49.285267 50.565468 50.565468 15190400
2004-09-03 50.286514 50.680038 49.474556 49.818268 49.818268 5176800
2004-09-07 50.316402 50.809555 49.619015 50.600338 50.600338 5875200
2004-09-08 50.181908 51.322632 50.062355 50.958992 50.958992 5009200
2004-09-09 51.073563 51.163227 50.311420 50.963974 50.963974 4080900
2004-09-10 50.610302 53.081039 50.460861 52.468334 52.468334 8740200
2004-09-13 53.115910 54.002586 53.031227 53.549286 53.549286 7881300
2004-09-14 53.524376 55.790882 53.195610 55.536835 55.536835 10880300
2004-09-15 55.073570 56.901718 54.894241 55.790882 55.790882 10763900
2004-09-16 55.960247 57.683788 55.616535 56.772205 56.772205 9310200
2004-09-17 56.996365 58.525631 56.562988 58.525631 58.525631 9517400
2004-09-20 58.256641 60.572956 58.166977 59.457142 59.457142 10679200
2004-09-21 59.681301 59.985161 58.535595 58.699978 58.699978 7263000
2004-09-22 58.480801 59.611561 58.186901 58.968971 58.968971 7617100
2004-09-23 59.198112 61.086033 58.291508 60.184414 60.184414 8576100
2004-09-24 60.244190 61.818291 59.656395 59.691261 59.691261 9166700
2004-09-27 59.556767 60.214302 58.680054 58.909195 58.909195 7099600
2004-09-28 60.423519 63.462128 59.880554 63.193138 63.193138 17009400
2004-09-29 63.113434 67.257904 62.879314 65.295258 65.295258 30661400
2004-09-30 64.707458 65.902977 64.259140 64.558022 64.558022 13823300
... ... ... ... ... ... ...
2020-05-19 1386.996948 1392.000000 1373.484985 1373.484985 1373.484985 1280600
2020-05-20 1389.579956 1410.420044 1387.250000 1406.719971 1406.719971 1655400
2020-05-21 1408.000000 1415.489990 1393.449951 1402.800049 1402.800049 1385000
2020-05-22 1396.709961 1412.760010 1391.829956 1410.420044 1410.420044 1309400
2020-05-26 1437.270020 1441.000000 1412.130005 1417.020020 1417.020020 2060600
2020-05-27 1417.250000 1421.739990 1391.290039 1417.839966 1417.839966 1685800
2020-05-28 1396.859985 1440.839966 1396.000000 1416.729980 1416.729980 1692200
2020-05-29 1416.939941 1432.569946 1413.349976 1428.920044 1428.920044 1838100
2020-06-01 1418.390015 1437.959961 1418.000000 1431.819946 1431.819946 1217100
2020-06-02 1430.550049 1439.609985 1418.829956 1439.219971 1439.219971 1278100
2020-06-03 1438.300049 1446.552002 1429.776978 1436.380005 1436.380005 1256200
2020-06-04 1430.400024 1438.959961 1404.729980 1412.180054 1412.180054 1484300
2020-06-05 1413.170044 1445.050049 1406.000000 1438.390015 1438.390015 1734900
2020-06-08 1422.339966 1447.989990 1422.339966 1446.609985 1446.609985 1404200
2020-06-09 1445.359985 1468.000000 1443.209961 1456.160034 1456.160034 1409200
2020-06-10 1459.540039 1474.259033 1456.270020 1465.849976 1465.849976 1525200
2020-06-11 1442.479980 1454.474976 1402.000000 1403.839966 1403.839966 1991300
2020-06-12 1428.489990 1437.000000 1386.020020 1413.180054 1413.180054 1944200
2020-06-15 1390.800049 1424.800049 1387.920044 1419.849976 1419.849976 1503900
2020-06-16 1445.219971 1455.020020 1425.900024 1442.719971 1442.719971 1709200
2020-06-17 1447.160034 1460.000000 1431.380005 1451.119995 1451.119995 1548300
2020-06-18 1449.160034 1451.410034 1427.010010 1435.959961 1435.959961 1581900
2020-06-19 1444.000000 1447.800049 1421.349976 1431.719971 1431.719971 3157900
2020-06-22 1429.000000 1452.750000 1423.209961 1451.859985 1451.859985 1542400
2020-06-23 1455.640015 1475.941040 1445.239990 1464.410034 1464.410034 1429800
2020-06-24 1461.510010 1475.420044 1429.750000 1431.969971 1431.969971 1756000
2020-06-25 1429.900024 1442.900024 1420.000000 1441.329956 1441.329956 1230500
2020-06-26 1431.390015 1433.449951 1351.989990 1359.900024 1359.900024 4267700
2020-06-29 1358.180054 1395.599976 1347.010010 1394.969971 1394.969971 1810200
2020-06-30 1390.439941 1418.650024 1383.959961 1413.609985 1413.609985 2041600
[3994 rows x 6 columns]
Note that is important to mention the start date correctly just to ensure we are collecting data. If we don't do that we might end up having some NaN variables that could affect the output of our training.
Now that we have the data that we want to use, we need to define what defines our traning and validation data. As stocks could vary depending on the dates, the function I have created requires 3 basic arguments:
- Ticker Symbol: GOOG
- Start Date: Date as to when they started, in this case, it was 2004-Aug-01.
- Validation Date: Date as to when we want the validation to be considered. In this case, we specify 2017-01-01 as our data point.
Note that you will need to have configured TensorFlow, Keras ,and a GPU in order to run the samples below.
In this exercise, I'm only interested in the closing price which is the standard benchmark regarding stocks or securities.
Below you can find the chart with the division we will create between Training Data and Validation Data:
Also, the histogram showing the distribution of the prices:
In order to normalise the data, we need to scale it between 0 and 1 so we talk on a common scale. To accomplish this, we can use the preprocessing tool MinMaxScaler as seen below:
min_max = MinMaxScaler(feature_range=(0, 1))
train_scaled = min_max.fit_transform(training_data)
To implement the LSTM network for time series forecasting, we have used a time step of 3 days. This technique allows the network to look back at the previous 3 days of data to predict the subsequent day. The figure below illustrates how this concept is used in our implementation, where the first 3 samples for Close price generate the 4th sample and so on. This generates a matrix of shape (3,1), with 3 representing the time steps and 1 representing the number of features (Close price).
# Define the number of time steps
time_steps = 3
# Loop through the data to create partitions
for i in range(time_steps, train_scaled.shape[0]):
# Create a partition of the previous 3 days' data
x_train.append(train_scaled[i - time_steps:i])
# Append the next day's Close price to the label array
y_train.append(train_scaled[i, 0])
Using a time step of 3 days allows the LSTM network to identify patterns in the data and make predictions based on those patterns. By using a sliding window approach, we can train the network to predict future stock prices based on historical data.
To create this model, you will need to have TensorFlow, TensorFlow-Gpu and Keras installed in order for this to run. The code for this model can be seen below and the explanation for each layer is also defined below:
def create_long_short_term_memory_model(x_train):
model = Sequential()
# 1st layer with Dropout regularisation
# * units = add 100 neurons is the dimensionality of the output space
# * return_sequences = True to stack LSTM layers so the next LSTM layer has a three-dimensional sequence input
# * input_shape => Shape of the training dataset
model.add(LSTM(units=100, return_sequences=True, input_shape=(x_train.shape[1], 1)))
# 20% of the layers will be dropped
model.add(Dropout(0.2))
# 2nd LSTM layer
# * units = add 50 neurons is the dimensionality of the output space
# * return_sequences = True to stack LSTM layers so the next LSTM layer has a three-dimensional sequence input
model.add(LSTM(units=50, return_sequences=True))
# 20% of the layers will be dropped
model.add(Dropout(0.2))
# 3rd LSTM layer
# * units = add 50 neurons is the dimensionality of the output space
# * return_sequences = True to stack LSTM layers so the next LSTM layer has a three-dimensional sequence input
model.add(LSTM(units=50, return_sequences=True))
# 50% of the layers will be dropped
model.add(Dropout(0.5))
# 4th LSTM layer
# * units = add 50 neurons is the dimensionality of the output space
model.add(LSTM(units=50))
# 50% of the layers will be dropped
model.add(Dropout(0.5))
# Dense layer that specifies an output of one unit
model.add(Dense(units=1))
model.summary()
tf.keras.utils.plot_model(model, to_file=os.path.join(project_folder, 'model_lstm.png'), show_shapes=True,
show_layer_names=True)
return model
The rendered model can be seen in the image below, producing a model with more than 100k trainable parameters.
Layer (type) Output Shape Param #
=================================================================
lstm_1 (LSTM) (None, 60, 100) 40800
_________________________________________________________________
dropout_1 (Dropout) (None, 60, 100) 0
_________________________________________________________________
lstm_2 (LSTM) (None, 60, 50) 30200
_________________________________________________________________
dropout_2 (Dropout) (None, 60, 50) 0
_________________________________________________________________
lstm_3 (LSTM) (None, 60, 50) 20200
_________________________________________________________________
dropout_3 (Dropout) (None, 60, 50) 0
_________________________________________________________________
lstm_4 (LSTM) (None, 50) 20200
_________________________________________________________________
dropout_4 (Dropout) (None, 50) 0
_________________________________________________________________
dense_1 (Dense) (None, 1) 51
=================================================================
Total params: 111,451
Trainable params: 111,451
Non-trainable params: 0
Once we have defined the model, we need to specify the metrics we want to use to track how well our model is behaving and also the kind of optimizer we want to use for our training. I have also defined the patience I want my model to have and what is the rule defined for it.
defined_metrics = [
tf.keras.metrics.MeanSquaredError(name='MSE')
]
callback = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=3, mode='min', verbose=1)
model.compile(optimizer='adam', loss='mean_squared_error', metrics=defined_metrics)
history = model.fit(x_train, y_train, epochs=epochs, batch_size=batch_size, validation_data=(x_test, y_test),
callbacks=[callback])
This model is slightly fined tuned to reach the lowest validation loss. In this example, we reach a validation loss of 0.14% with an MSE (Mean Square Error) of 0.14% which is relatively good, providing us with a very accurate result.
The training result can be seen below:
Train on 3055 samples, validate on 881 samples
Epoch 1/100
2020-07-11 15:15:34.557035: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_100.dll
3112/3112 [==============================] - 19s 6ms/sample - loss: 0.0451 - MSE: 0.0451 - val_loss: 0.0068 - val_MSE: 0.0068
Epoch 2/100
3112/3112 [==============================] - 4s 1ms/sample - loss: 0.0088 - MSE: 0.0088 - val_loss: 0.0045 - val_MSE: 0.0045
Epoch 3/100
3112/3112 [==============================] - 5s 1ms/sample - loss: 0.0062 - MSE: 0.0062 - val_loss: 0.0032 - val_MSE: 0.0032
Epoch 4/100
3112/3112 [==============================] - 5s 1ms/sample - loss: 0.0051 - MSE: 0.0051 - val_loss: 0.0015 - val_MSE: 0.0015
Epoch 5/100
3112/3112 [==============================] - 7s 2ms/sample - loss: 0.0045 - MSE: 0.0045 - val_loss: 0.0013 - val_MSE: 0.0013
Epoch 6/100
3112/3112 [==============================] - 5s 2ms/sample - loss: 0.0045 - MSE: 0.0045 - val_loss: 0.0013 - val_MSE: 0.0013
Epoch 7/100
3112/3112 [==============================] - 5s 2ms/sample - loss: 0.0045 - MSE: 0.0045 - val_loss: 0.0015 - val_MSE: 0.0015
Epoch 8/100
3112/3112 [==============================] - 5s 1ms/sample - loss: 0.0040 - MSE: 0.0040 - val_loss: 0.0015 - val_MSE: 0.0015
Epoch 9/100
3112/3112 [==============================] - 5s 1ms/sample - loss: 0.0039 - MSE: 0.0039 - val_loss: 0.0014 - val_MSE: 0.0014
Epoch 00009: early stopping
saving weights
plotting loss
plotting MSE
display the content of the model
886/1 - 0s - loss: 0.0029 - MSE: 0.0014
loss : 0.0014113364930413916
MSE : 0.0014113366
Now it is time to prepare our testing data and send it through our deep-learning model to obtain the predictions we are trying to get.
First, we need to import the test data using the same approach we used for the training data using the time steps:
def infer_data():
print(tf.version.VERSION)
inference_folder = os.path.join(os.getcwd(), RUN_FOLDER)
stock = StockPrediction(STOCK_TICKER, STOCK_START_DATE, STOCK_VALIDATION_DATE, inference_folder, GITHUB_URL, EPOCHS, TIME_STEPS, TOKEN, BATCH_SIZE)
data = StockData(stock)
(x_train, y_train), (x_test, y_test), (training_data, test_data) = data.download_transform_to_numpy(TIME_STEPS, inference_folder)
min_max = data.get_min_max()
# load future data
print('Latest Stock Price')
latest_close_price = test_data.Close.iloc[-1]
latest_date = test_data[-1:]['Close'].idxmin()
print(latest_close_price)
print('Latest Date')
print(latest_date)
tomorrow_date = latest_date + timedelta(1)
# Specify the next 300 days
next_year = latest_date + timedelta(TIME_STEPS * 100)
print('Future Date')
print(tomorrow_date)
print('Future Timespan Date')
print(next_year)
x_test, y_test, test_data = data.generate_future_data(TIME_STEPS, min_max, tomorrow_date, next_year, latest_close_price)
# Check if the future data is not empty
if x_test.shape[0] > 0:
# load the weights from our best model
model = tf.keras.models.load_model(os.path.join(inference_folder, 'model_weights.h5'))
model.summary()
# perform a prediction
test_predictions_baseline = model.predict(x_test)
test_predictions_baseline = min_max.inverse_transform(test_predictions_baseline)
test_predictions_baseline = pd.DataFrame(test_predictions_baseline, columns=['Predicted_Price'])
# Combine the predicted values with dates from the test data
predicted_dates = pd.date_range(start=test_data.index[0], periods=len(test_predictions_baseline))
test_predictions_baseline['Date'] = predicted_dates
# Reset the index for proper concatenation
test_data.reset_index(inplace=True)
# Concatenate the test_data and predicted data
combined_data = pd.concat([test_data, test_predictions_baseline], ignore_index=True)
# Plotting predictions
plt.figure(figsize=(14, 5))
plt.plot(combined_data['Date'], combined_data.Close, color='green', label='Simulated [' + STOCK_TICKER + '] price')
plt.plot(combined_data['Date'], combined_data['Predicted_Price'], color='red', label='Predicted [' + STOCK_TICKER + '] price')
plt.xlabel('Time')
plt.ylabel('Price [USD]')
plt.legend()
plt.title('Simulated vs Predicted Prices')
plt.savefig(os.path.join(inference_folder, STOCK_TICKER + '_future_comparison.png'))
plt.show()
else:
print("Error: Future data is empty.")
start_date = pd.to_datetime('2017-01-01')
end_date = datetime.today()
duration = end_date - start_date
STOCK_VALIDATION_DATE = start_date + 0.8 * duration
Now we can call the infer_data
method which will allow us to generate the stock prediction based on the training done over the training data. As a result, we will generate a csv file that contains the result of the prediction and also a chart that shows what's the real vs the estimation.
This has been built using Python 3.10 version.
Download the source code and install the following packages:
C:\Users\thund\Source\Repos\stock-prediction-deep-neural-learning>pip list
Package Version
-------------------- ---------
absl-py 0.13.0
alabaster 0.7.12
anaconda-client 1.7.2
anaconda-navigator 2.0.3
anaconda-project 0.9.1
anyio 2.2.0
appdirs 1.4.4
argh 0.26.2
argon2-cffi 20.1.0
asn1crypto 1.4.0
astroid 2.6.6
astropy 4.2.1
astunparse 1.6.3
async-generator 1.10
atomicwrites 1.4.0
attrs 20.3.0
autopep8 1.5.5
Babel 2.9.0
backcall 0.2.0
backports.functools-lru-cache 1.6.4
backports.shutil-get-terminal-size 1.0.0
backports.tempfile 1.0
backports.weakref 1.0.post1
bcrypt 3.2.0
beautifulsoup4 4.9.3
bitarray 1.9.2
bkcharts 0.2
black 19.10b0
bleach 3.3.0
bokeh 2.3.2
boto 2.49.0
Bottleneck 1.3.2
brotlipy 0.7.0
cachetools 4.2.2
certifi 2020.12.5
cffi 1.14.5
chardet 4.0.0
clang 5.0
click 7.1.2
cloudpickle 1.6.0
clyent 1.2.2
colorama 0.4.4
comtypes 1.1.9
conda 4.10.1
conda-build 3.21.4
conda-content-trust 0+unknown
conda-package-handling 1.7.3
conda-repo-cli 1.0.4
conda-token 0.3.0
conda-verify 3.4.2
contextlib2 0.6.0.post1
cryptography 3.4.7
cycler 0.10.0
Cython 0.29.23
cytoolz 0.11.0
dask 2021.4.0
decorator 5.0.6
defusedxml 0.7.1
diff-match-patch 20200713
distributed 2021.4.0
docutils 0.17
entrypoints 0.3
et-xmlfile 1.0.1
fastcache 1.1.0
filelock 3.0.12
flake8 3.8.4
Flask 1.1.2
flatbuffers 1.12
fsspec 0.9.0
future 0.18.2
gast 0.4.0
gevent 21.1.2
glob2 0.7
google-auth 1.35.0
google-auth-oauthlib 0.4.6
google-pasta 0.2.0
greenlet 1.0.0
grpcio 1.40.0
h5py 3.1.0
HeapDict 1.0.1
helpdev 0.7.1
html5lib 1.1
idna 2.10
imagecodecs 2021.3.31
imageio 2.9.0
imagesize 1.2.0
importlib-metadata 3.10.0
iniconfig 1.1.1
intervaltree 3.1.0
ipykernel 5.3.4
ipython 7.22.0
ipython-genutils 0.2.0
ipywidgets 7.6.3
isort 5.8.0
itsdangerous 1.1.0
jdcal 1.4.1
jedi 0.17.2
Jinja2 2.11.3
joblib 1.0.1
json5 0.9.5
jsonschema 3.2.0
jupyter 1.0.0
jupyter-client 6.1.12
jupyter-console 6.4.0
jupyter-core 4.7.1
jupyter-packaging 0.7.12
jupyter-server 1.4.1
jupyterlab 3.0.14
jupyterlab-pygments 0.1.2
jupyterlab-server 2.4.0
jupyterlab-widgets 1.0.0
keras 2.6.0
Keras-Preprocessing 1.1.2
keyring 22.3.0
kiwisolver 1.3.1
lazy-object-proxy 1.6.0
libarchive-c 2.9
llvmlite 0.36.0
locket 0.2.1
lxml 4.6.3
Markdown 3.3.4
MarkupSafe 1.1.1
matplotlib 3.3.4
mccabe 0.6.1
menuinst 1.4.16
mistune 0.8.4
mkl-fft 1.3.0
mkl-random 1.2.1
mkl-service 2.3.0
mock 4.0.3
more-itertools 8.7.0
mpmath 1.2.1
msgpack 1.0.2
multipledispatch 0.6.0
multitasking 0.0.9
mypy-extensions 0.4.3
navigator-updater 0.2.1
nbclassic 0.2.6
nbclient 0.5.3
nbconvert 6.0.7
nbformat 5.1.3
nest-asyncio 1.5.1
networkx 2.5
nltk 3.6.1
nose 1.3.7
notebook 6.3.0
numba 0.53.1
numexpr 2.7.3
numpy 1.21.2
numpydoc 1.1.0
oauthlib 3.1.1
olefile 0.46
openpyxl 3.0.7
opt-einsum 3.3.0
packaging 20.9
pandas 1.2.4
pandocfilters 1.4.3
paramiko 2.7.2
parso 0.7.0
partd 1.2.0
path 15.1.2
pathlib2 2.3.5
pathspec 0.7.0
patsy 0.5.1
pep8 1.7.1
pexpect 4.8.0
pickleshare 0.7.5
Pillow 8.2.0
pip 21.2.4
pkginfo 1.7.0
pluggy 0.13.1
ply 3.11
prometheus-client 0.10.1
prompt-toolkit 3.0.17
protobuf 3.18.0
psutil 5.8.0
ptyprocess 0.7.0
py 1.10.0
pyasn1 0.4.8
pyasn1-modules 0.2.8
pycodestyle 2.6.0
pycosat 0.6.3
pycparser 2.20
pycurl 7.43.0.6
pydocstyle 6.0.0
pyerfa 1.7.3
pyflakes 2.2.0
Pygments 2.8.1
pylint 2.7.4
pyls-black 0.4.6
pyls-spyder 0.3.2
PyNaCl 1.4.0
pyodbc 4.0.0-unsupported
pyOpenSSL 20.0.1
pyparsing 2.4.7
PyQt5 5.12.3
PyQt5-sip 12.9.0
PyQtWebEngine 5.12.1
pyreadline 2.1
pyrsistent 0.17.3
PySocks 1.7.1
pytest 6.2.3
python-dateutil 2.8.1
python-jsonrpc-server 0.4.0
python-language-server 0.36.2
pytz 2021.1
PyWavelets 1.1.1
pywin32 227
pywin32-ctypes 0.2.0
pywinpty 0.5.7
PyYAML 5.4.1
pyzmq 20.0.0
QDarkStyle 2.8.1
QtAwesome 1.0.2
qtconsole 5.0.3
QtPy 1.9.0
regex 2021.4.4
requests 2.25.1
requests-oauthlib 1.3.0
rope 0.18.0
rsa 4.7.2
Rtree 0.9.7
ruamel-yaml-conda 0.15.100
scikit-image 0.18.1
scikit-learn 0.24.1
scipy 1.6.2
seaborn 0.11.1
Send2Trash 1.5.0
setuptools 52.0.0.post20210125
simplegeneric 0.8.1
singledispatch 0.0.0
sip 4.19.13
six 1.15.0
sniffio 1.2.0
snowballstemmer 2.1.0
sortedcollections 2.1.0
sortedcontainers 2.3.0
soupsieve 2.2.1
Sphinx 4.0.1
sphinxcontrib-applehelp 1.0.2
sphinxcontrib-devhelp 1.0.2
sphinxcontrib-htmlhelp 1.0.3
sphinxcontrib-jsmath 1.0.1
sphinxcontrib-qthelp 1.0.3
sphinxcontrib-serializinghtml 1.1.4
sphinxcontrib-websupport 1.2.4
spyder 4.2.5
spyder-kernels 1.10.2
SQLAlchemy 1.4.7
statsmodels 0.12.2
sympy 1.8
tables 3.6.1
tblib 1.7.0
tensorboard 2.6.0
tensorboard-data-server 0.6.1
tensorboard-plugin-wit 1.8.0
tensorflow 2.6.0
tensorflow-estimator 2.6.0
termcolor 1.1.0
terminado 0.9.4
testpath 0.4.4
textdistance 4.2.1
threadpoolctl 2.1.0
three-merge 0.1.1
tifffile 2021.4.8
toml 0.10.2
toolz 0.11.1
tornado 6.1
tqdm 4.59.0
traitlets 5.0.5
typed-ast 1.4.2
typing-extensions 3.7.4.3
ujson 4.0.2
unicodecsv 0.14.1
urllib3 1.26.4
watchdog 1.0.2
wcwidth 0.2.5
webencodings 0.5.1
Werkzeug 1.0.1
wheel 0.36.2
widgetsnbextension 3.5.1
win-inet-pton 1.1.0
win-unicode-console 0.5
wincertstore 0.2
wrapt 1.12.1
xlrd 2.0.1
XlsxWriter 1.3.8
xlwings 0.23.0
xlwt 1.3.0
xmltodict 0.12.0
yapf 0.31.0
yfinance 0.2.33
zict 2.0.0
zipp 3.4.1
zope.event 4.5.0
zope.interface 5.3.0
Then edit the file "stock_prediction_deep_learning.py" to include the Stock you want to use and the relevant dates and execute:
# all arguments are optional
python stock_prediction_deep_learning.py -ticker=GOOG -start_date=2017-11-01 -validation_date=2022-09-01 -epochs=100 -batch_size=32 -time_steps=3
Also you can use the Jupyter notebook
available here:
As I mentioned before, I'm using a GPU to ramp up my testing. As my laptop has a nvidia geforce card, I installed CUDA to make use of its GPU capabilities. Depending on your tensorflow version you'll need a version or another. Here is the link: https://developer.nvidia.com/cuda-11.0-download-archive?target_os=Windows&target_arch=x86_64&target_version=10&target_type=exelocal
You can do from your conda prompt:
(base) >conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Collecting package metadata (repodata.json): done
Solving environment: done
==> WARNING: A newer version of conda exists. <==
current version: 4.10.3
latest version: 4.14.0
Please update conda by running
$ conda update -n base -c defaults conda
## Package Plan ##
environment location: C:\Users\jordi\anaconda3
added / updated specs:
- cudatoolkit=11.2
- cudnn=8.1.0
The following packages will be downloaded:
package | build
---------------------------|-----------------
conda-4.12.0 | py39hcbf5309_0 1.0 MB conda-forge
cudatoolkit-11.2.2 | h933977f_10 879.9 MB conda-forge
cudnn-8.1.0.77 | h3e0f4f4_0 610.8 MB conda-forge
python_abi-3.9 | 2_cp39 4 KB conda-forge
------------------------------------------------------------
Total: 1.46 GB
The following NEW packages will be INSTALLED:
cudatoolkit conda-forge/win-64::cudatoolkit-11.2.2-h933977f_10
cudnn conda-forge/win-64::cudnn-8.1.0.77-h3e0f4f4_0
python_abi conda-forge/win-64::python_abi-3.9-2_cp39
The following packages will be UPDATED:
conda pkgs/main::conda-4.10.3-py39haa95532_0 --> conda-forge::conda-4.12.0-py39hcbf5309_0
Proceed ([y]/n)? y
Downloading and Extracting Packages
conda-4.12.0 | 1.0 MB | ############################################################################ | 100%
python_abi-3.9 | 4 KB | ############################################################################ | 100%
cudnn-8.1.0.77 | 610.8 MB | ############################################################################ | 100%
cudatoolkit-11.2.2 | 879.9 MB | ############################################################################ | 100%
Preparing transaction: done
Verifying transaction: done
Executing transaction: / "By downloading and using the CUDA Toolkit conda packages, you accept the terms and conditions of the CUDA End User License Agreement (EULA): https://docs.nvidia.com/cuda/eula/index.html"
| "By downloading and using the cuDNN conda packages, you accept the terms and conditions of the NVIDIA cuDNN EULA - https://docs.nvidia.com/deeplearning/cudnn/sla/index.html"
done
If you run the project after this, the GPU should be correctly picked up:
if you see this message, then you need to install GraphViz library:
You must install pydot (`pip install pydot`) and install graphviz (see instructions at https://graphviz.gitlab.io/download/) for plot_model/model_to_dot to work.
No sponsors yet! Will you be the first?