Upload main.py
Browse files
main.py
ADDED
@@ -0,0 +1,126 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# -*- coding: utf-8 -*-
|
2 |
+
"""
|
3 |
+
Created on Sun Nov 6 18:29:53 2022
|
4 |
+
|
5 |
+
@author: culli
|
6 |
+
"""
|
7 |
+
|
8 |
+
import pandas as pd
|
9 |
+
import numpy as np
|
10 |
+
import pandas_datareader.data as web
|
11 |
+
import datetime as dt
|
12 |
+
from datetime import timedelta
|
13 |
+
from sklearn.preprocessing import MinMaxScaler
|
14 |
+
import tensorflow as tf
|
15 |
+
from tensorflow import lite
|
16 |
+
from tensorflow.keras.models import Sequential
|
17 |
+
from tensorflow.keras.layers import Dense, Dropout, LSTM
|
18 |
+
import matplotlib.pyplot as plt
|
19 |
+
|
20 |
+
# Load Training Data
|
21 |
+
company = "BTC-USD"
|
22 |
+
start = dt.datetime(2015,1,1)
|
23 |
+
end = dt.datetime(2022,1,1)
|
24 |
+
data = web.DataReader(company, "yahoo", start, end)
|
25 |
+
|
26 |
+
|
27 |
+
# Prepare Data
|
28 |
+
scaler = MinMaxScaler(feature_range=(0,1))
|
29 |
+
scaled_data = scaler.fit_transform(data['Close'].values.reshape(-1,1))
|
30 |
+
|
31 |
+
prediction_days = 60
|
32 |
+
|
33 |
+
x_train = []
|
34 |
+
y_train = []
|
35 |
+
|
36 |
+
for x in range(prediction_days, len(scaled_data)):
|
37 |
+
x_train.append(scaled_data[x-prediction_days:x, 0])
|
38 |
+
y_train.append(scaled_data[x, 0])
|
39 |
+
|
40 |
+
|
41 |
+
x_train, y_train = np.array(x_train), np.array(y_train)
|
42 |
+
x_train = np.reshape(x_train, (x_train.shape[0], x_train.shape[1], 1))
|
43 |
+
|
44 |
+
|
45 |
+
# Build the model
|
46 |
+
model = Sequential()
|
47 |
+
|
48 |
+
model.add(LSTM(units=50, return_sequences=True, input_shape = (x_train.shape[1], 1)))
|
49 |
+
model.add(Dropout(0.2))
|
50 |
+
model.add(LSTM(units=50, return_sequences=True))
|
51 |
+
model.add(Dropout(0.2))
|
52 |
+
model.add(LSTM(units=50))
|
53 |
+
model.add(Dropout(0.2))
|
54 |
+
model.add(Dense(units=1))
|
55 |
+
|
56 |
+
|
57 |
+
model.compile(optimizer="adam", loss="mean_squared_error")
|
58 |
+
model.fit(x_train, y_train, epochs=25, batch_size=64)
|
59 |
+
|
60 |
+
# Load Data
|
61 |
+
test_start = dt.datetime(2022,1,1)
|
62 |
+
test_end = dt.datetime.now()
|
63 |
+
|
64 |
+
test_data = web.DataReader(company, "yahoo", test_start, test_end)
|
65 |
+
actual_prices = test_data["Close"].values
|
66 |
+
|
67 |
+
total_dataset = pd.concat((data["Close"], test_data["Close"]), axis = 0)
|
68 |
+
|
69 |
+
|
70 |
+
|
71 |
+
model_inputs = total_dataset[len(total_dataset) - len(test_data) - prediction_days:].values # Important
|
72 |
+
model_inputs = model_inputs.reshape(-1, 1)
|
73 |
+
model_inputs = scaler.transform(model_inputs)
|
74 |
+
|
75 |
+
# Loop for making predictions
|
76 |
+
for x in range(prediction_days):
|
77 |
+
|
78 |
+
|
79 |
+
# Make Predictions on Test Data
|
80 |
+
x_test = []
|
81 |
+
|
82 |
+
for x in range(prediction_days, len(model_inputs)+1):
|
83 |
+
x_test.append(model_inputs[x-prediction_days:x, 0])
|
84 |
+
|
85 |
+
x_test = np.array(x_test)
|
86 |
+
x_test = np.reshape(x_test, (x_test.shape[0], x_test.shape[1], 1))
|
87 |
+
|
88 |
+
predicted_prices = model.predict(x_test)
|
89 |
+
predicted_prices = scaler.inverse_transform(predicted_prices)
|
90 |
+
|
91 |
+
|
92 |
+
# Predict Next Day
|
93 |
+
real_data = [model_inputs[len(model_inputs) + 1 - prediction_days:len(model_inputs+1), 0]]
|
94 |
+
real_data = np.array(real_data)
|
95 |
+
real_data = np.reshape(real_data, (real_data.shape[0], real_data.shape[1], 1))
|
96 |
+
|
97 |
+
# Formatting the prediction and printing it
|
98 |
+
prediction = model.predict(real_data)
|
99 |
+
prediction = scaler.inverse_transform(prediction)
|
100 |
+
print(f"Prediction: {prediction}")
|
101 |
+
|
102 |
+
# Adding the prediction back into the model for the next prediction
|
103 |
+
model_prediction_input = prediction
|
104 |
+
model_prediction_input = model_prediction_input.reshape(-1, 1)
|
105 |
+
model_prediction_input = scaler.transform(model_prediction_input)
|
106 |
+
model_inputs = np.concatenate((model_inputs, model_prediction_input))
|
107 |
+
|
108 |
+
|
109 |
+
plt.plot(actual_prices, color="black", label=f"Actual {company} Price")
|
110 |
+
plt.plot(predicted_prices, color="green", label=f"Predicted {company} Price")
|
111 |
+
plt.title(f"{company} Share Price")
|
112 |
+
plt.xlabel("Days")
|
113 |
+
plt.ylabel(f"{company} Share Price")
|
114 |
+
plt.legend()
|
115 |
+
plt.show()
|
116 |
+
|
117 |
+
|
118 |
+
tf.keras.models.save_model(model,"model.pbtxt")
|
119 |
+
converter = lite.TFLiteConverter.from_keras_model(model = model)
|
120 |
+
converter.optimizations = [tf.lite.Optimize.DEFAULT]
|
121 |
+
converter.experimental_new_converter=True
|
122 |
+
converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS,
|
123 |
+
tf.lite.OpsSet.SELECT_TF_OPS]
|
124 |
+
model_tflite = converter.convert()
|
125 |
+
open("BTCPrediction.tflite", "wb").write(model_tflite)
|
126 |
+
|