Skip to content

Commit

Permalink
Resolved conflict
Browse files Browse the repository at this point in the history
  • Loading branch information
@rguptauw
rguptauw committed Jun 4, 2018
1 parent 6af13fa commit 8f40e92
Showing 1 changed file with 110 additions and 29 deletions.
139 changes: 110 additions & 29 deletions pybcoin/ModelForecast/btc_model.py
View file
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
"""
Module name: btc_model
Description: Module for predicting bitcoin price direction based on
external regressors and bitcoin time series.
external regressors and time series.
"""

# importing modules
Expand Down Expand Up @@ -34,13 +34,17 @@ class BtcModelPrediction(object):

# constructor function
def __init__(self, params):
self.in_path_btc = params['Forecast']['in_path_btc']
self.in_path_comm = params['Forecast']['in_path_comm']
self.in_path_gtrends = params['Forecast']['in_path_gtrends']
self.out_path = params['Forecast']['out_path']

self.in_path_btc = params['in_path_btc']
self.in_path_comm = params['in_path_comm']
self.in_path_gtrends = params['in_path_gtrends']
self.in_path_social = params['in_path_social']
self.path_time_pred = params['path_time_pred']
self.out_path = params['out_path']

# function to create a time series prediction
def time_prediction(self):

"""
Method to predict the BTC price for one day based on historical
prices. Uses a library 'prophet' made open-source by Facebook.
Expand All @@ -63,29 +67,43 @@ def time_prediction(self):
# training from fbprophet
# prophet module requires columns ds (date) and y (value)
candles = btc_data.rename(columns={'utc_time': 'ds',
'btc_price': 'y'})
'btc_price': 'y'})
m = Prophet(yearly_seasonality=True, daily_seasonality=False,
changepoint_prior_scale=0.001)
m.fit(candles)

# making the predictions
future = m.make_future_dataframe(periods=1, include_history=False)
future = m.predict(future)
ts_res = future['yhat'][0]
return ts_res

# creating the row to append to the time series dataset
today_date = datetime.datetime.strptime(self.today_date,
'%Y-%m-%d')
pred_date = today_date + datetime.timedelta(days=1)
pred_date = pred_date.strftime("%m/%d/%Y")

app_list = [pred_date]
app_list.append(future['yhat'][0])

# writing to the file
with open(self.path_time_pred + 'predicted_time.csv', 'a') as f:
writer = csv.writer(f)
writer.writerow(app_list)

return future['yhat']

except Exception as e:
print(e)
return error_val

# function to create a linear regression model
def linear_prediction(self):

"""
Method to predict BTC price for next day on external regressors:
1. Oil Price
2. Google trends
Uses scikit learn library to run linear regression.
: param self
: return pred_reg(float)
error_val(int)
Expand All @@ -95,34 +113,89 @@ def linear_prediction(self):
oil_data = pd.read_csv(self.in_path_comm + 'oil_price.csv')
google_data = pd.read_csv(self.in_path_gtrends + 'gtrends.csv')
btc_data = pd.read_csv(self.in_path_btc + 'btc_price.csv')
twitter_sentiment = pd.read_csv(self.in_path_social +
'twitter_sentiment.csv')
reddit_sentiment = pd.read_csv(self.in_path_social +
'reddit_sentiment.csv')

# merging all the datasets
df = pd.merge(btc_data, oil_data, left_on='utc_time',
right_on='Date')
del df['Date']
df = pd.merge(df, google_data, left_on='utc_time',
right_on='date')
df = pd.merge(btc_data, oil_data,
left_on='utc_time', right_on='dates_s')
del df['dates_s']
df = pd.merge(df, google_data, left_on='utc_time', right_on='date')
del df['date']
df['google_hits'] = df['btc'] + df['bitcoin'] + df['btc usd']
+ df['btcusd']
df = pd.merge(df, twitter_sentiment,
left_on='utc_time', right_on='Date')
del df['Date']

# renaming columns name for twitter
df = df.rename(columns={'Negative': 'twitter_negative'})
df = df.rename(columns={'Positive': 'twitter_positive'})
df = pd.merge(df, reddit_sentiment,
left_on='utc_time', right_on='Date')
del df['Date']

# renaming columns name for reddit
df = df.rename(columns={'Negative': 'reddit_negative'})
df = df.rename(columns={'Positive': 'reddit_positive'})

# getting relevant computations from social data
df['google_hits'] = (df['btc'] + df['bitcoin'] +
df['btc usd'] + df['btcusd'])
df['twitter'] = (df['twitter_positive'] / (df['twitter_positive'] +
df['twitter_negative']))
df['reddit'] = (df['reddit_positive'] / (df['reddit_positive'] +
df['reddit_negative']))

# selecting only relevant columns
df = df[['utc_time', 'oil_price', 'google_hits', 'btc_price']]
df = df[['utc_time', 'oil_price', 'google_hits', 'twitter',
'reddit', 'btc_price']]

# initializing the columns of price deltas
df['btc_delta'] = 0
df['oil_delta'] = 0
df['google_delta'] = 0
df['twitter_delta'] = 0
df['reddit_delta'] = 0

# computing the percentage change for all rows
for i in range(1, len(df)):
df['btc_delta'].iloc[i] = ((df['btc_price'].iloc[i] -
df['btc_price'].iloc[i - 1]) /
df['btc_price'].iloc[i - 1])
df['oil_delta'].iloc[i] = ((df['oil_price'].iloc[i] -
df['oil_price'].iloc[i - 1]) /
df['oil_price'].iloc[i - 1])
df['google_delta'].iloc[i] = ((df['google_hits'].iloc[i] -
df['google_hits'].iloc[i - 1]) /
df['google_hits'].iloc[i - 1])
df['twitter_delta'].iloc[i] = ((df['twitter'].iloc[i] -
df['twitter'].iloc[i - 1]) /
df['twitter'].iloc[i - 1])
df['reddit_delta'].iloc[i] = ((df['reddit'].iloc[i] -
df['reddit'].iloc[i - 1]) /
df['reddit'].iloc[i - 1])

# shifting the btc price by +1 day, want to predict for next day
df['btc_price_next'] = 0
# shifting the btc price by +1 day, want to predict for next day
df['btc_delta_pred'] = 0
for i in range(0, len(df) - 1):
df['btc_price_next'].iloc[i] = df['btc_price'].iloc[i + 1]
df['btc_delta_pred'].iloc[i] = df['btc_delta'].iloc[i + 1]

# selecting only relevant columns
df = df[['utc_time', 'oil_delta', 'google_delta', 'twitter_delta',
'reddit_delta', 'btc_price', 'btc_delta',
'btc_delta_pred']]

# preparing the dataset for linear regression
df_lin = df[['oil_price', 'google_hits', 'btc_price_next']]
df_lin = df[['oil_delta', 'google_delta', 'twitter_delta',
'reddit_delta', 'btc_delta_pred']]

# response index number
resp_idx = df_lin.shape[1] - 1

# last row are the new features with which prediction is to be made
feature_day = df_lin.iloc[-1]
del feature_day['btc_price_next']
del feature_day['btc_delta_pred']
X_day = feature_day.as_matrix()
X_day = X_day.reshape(1, -1)

Expand All @@ -138,6 +211,7 @@ def linear_prediction(self):
regressor.fit(X, y)
pred_reg = regressor.predict(X_day)[0]
return pred_reg

except Exception as e:
print(e)
return error_val
Expand All @@ -153,14 +227,20 @@ def final_prediction(self):
"""
error_val = -1
try:
final_yhat = (self.time_prediction() +
self.linear_prediction()) / 2
time_data = pd.read_csv(self.path_time_pred + 'predicted_time.csv')
price_old = time_data['predicted_price_time'].iloc[-2]
time_ratio = (self.time_prediction() - price_old) / price_old
reg_ratio = self.linear_prediction()

final_yhat = (reg_ratio * 0.8) + (time_ratio * 0.2)

# predicted direction
res_dir = final_yhat - self.curr_price
res = 1 if np.sign(res_dir) == 1.0 else 0
res = 1 if np.sign(final_yhat)[0] == 1.0 else 0

# confidence calculation
ci = np.absolute(res_dir) / self.curr_price
ci_f = round(ci * 10, 2)
ci = round((np.absolute(final_yhat[0]) / 0.05), 2)
ci_f = ci if ci < 1 else 1

# creating the row to append finally
today_date = datetime.datetime.strptime(self.today_date,
'%Y-%m-%d')
Expand All @@ -172,9 +252,10 @@ def final_prediction(self):
app_list.append(ci_f)

# writing to the file
with open(self.out_path + 'BitcoinPrice.csv', 'ab') as f:
with open(self.out_path + 'BitcoinPrice.csv', 'a') as f:
writer = csv.writer(f)
writer.writerow(app_list)

except Exception as e:
print(e)
return error_val

0 comments on commit 8f40e92

Please sign in to comment.