def test_next_batch_production(self):
rec_ts = TimeSeries(self.input_data)
X_feature_space, y_target = rec_ts.next_batch(batch_size=4, n_steps=10)
self.assertEqual(len(X_feature_space), 4)
self.assertEqual(len(X_feature_space[0]), 10)
self.assertEqual(len(X_feature_space[0][0]), 2)
self.assertEqual(X_feature_space[3][0][0], y_target[3][0][0])
def test_next_batch_covariates(self):
"""
Feature space is supplied in input if target_only is False (no need to lag y dataset)
"""
rec_ts = TimeSeries(self.input_data)
X_feature_space, y_target = rec_ts.next_batch(batch_size=1, n_steps=10)
self.assertEqual(len(X_feature_space), 1)
self.assertEqual(len(X_feature_space[0][0]), 2)
def test_sample_ts(self):
"""
When the length of the pandas df is longer than required length the function should sample
from the time series and return that sample
"""
rec_instance = TimeSeries(pandas_df=self.data_to_pad)
results = rec_instance._sample_ts(pandas_df=self.data_to_pad,
desired_len=3)
self.assertEqual(results.shape[0], 3)
def test_padding_with_one_hot(self):
rec_ts = TimeSeries(pandas_df=self.data_to_pad_with_categorical,
one_hot_root_list=["one_hot"])
results = rec_ts._pad_ts(pandas_df=self.data_to_pad_with_categorical,
desired_len=10)
self.assertEqual(results.shape[0], 10)
self.assertEqual(results.one_hot_yes.values[0], 1)
self.assertEqual(results.one_hot_no.values[0], 0)
def test_next_batch_covariates_3(self):
"""
Feature space is supplied in input if target_only is False (no need to lag y dataset)
"""
rec_ts = TimeSeries(self.input_data)
X_feature_space, y_target = rec_ts.next_batch(batch_size=2, n_steps=20)
print('X_feature_space:', X_feature_space.shape, X_feature_space)
print('y_target:', y_target.shape, y_target)
self.assertEqual(len(X_feature_space), 2)
self.assertEqual(len(X_feature_space[0][0]), 2)
def _create_ts_object(df, dataset):
if dataset == '56_sunspots':
ds = TimeSeries(df, target_idx=4, timestamp_idx=1, index_col=0)
elif dataset == 'LL1_736_population_spawn':
ds = TimeSeries(df,
target_idx=2,
timestamp_idx=1,
index_col=0,
grouping_idx=3,
count_data=True)
return ds
def test_zero_len_padding(self):
rec_instance = TimeSeries(pandas_df=self.data_to_pad)
results = rec_instance._pad_ts(
pandas_df=self.data_to_pad,
desired_len=6) # len is the same as the original time series
self.assertEqual(results.shape[0], 6)
def test_len_padding(self):
rec_instance = TimeSeries(pandas_df=self.data_to_pad)
results = rec_instance._pad_ts(pandas_df=self.data_to_pad,
desired_len=10)
self.assertEqual(results.shape[0], 10)
from deepar.model.loss import gaussian_likelihood,gaussian_likelihood_2
import os
data = pd.read_csv(os.getcwd()+'\data\%s' % 'B007SIR08C-A23TNQB4GVF91M-ATVPDKIKX0DER-1.csv', header=None, names=['date','order','seller','marketplace'])
data['count'] = data['date'].apply(lambda x: int(x.split('-')[0])*10000+int(x.split('-')[1])*100+int(x.split('-')[2]))
data['promotion'] = 0
data.loc[data['order'] > 150, 'promotion'] = 1
order_max = data['order'].max()
order_min = data['order'].min()
data[['order','count']] = data[['order','count']].apply(lambda x : (x-np.min(x))/(np.max(x)-np.min(x)))
data['date'] = pd.to_datetime(data['date'])
data.set_index('date', inplace=True)
data.drop(columns=['seller','marketplace'], inplace=True)
ts = TimeSeries(data.head(500))
# ts = MockTs()
dp_model = DeepAR(ts, epochs=100)
dp_model.init()
dp_model.model.load_weights('1.h5', by_name=True)
# dp_model.more_fit()
def sigmoid(x):
y = 1/(1+np.exp(-1*x))
return y
def tanh(x):
y = 2*sigmoid(2*x)-1
import pandas as pd
import numpy as np
from matplotlib import pyplot as plt
import os
file_list = os.listdir('data')
for file_path in file_list:
data = pd.read_csv(os.getcwd() + '\data\%s' % file_path,
header=None,
names=['date', 'order', 'seller', 'marketplace'])
data['date'] = pd.to_datetime(data['date'])
data.set_index('date', inplace=True)
data.drop(columns=['seller', 'marketplace'], inplace=True)
ts = TimeSeries(data.head(265))
# ts = MockTs()
dp_model = DeepAR(ts, epochs=150)
dp_model.instantiate_and_fit()
def get_sample_prediction(sample, fn):
sample = np.array(sample).reshape(1, 30, 1)
output = fn([sample])
samples = []
# return output[0].reshape(1)
for mu, sigma in zip(output[0].reshape(1), output[1].reshape(1)):
samples.append(normal(loc=mu, scale=np.sqrt(sigma), size=1)[0])
return np.array(samples)
# predict_data = ts.next_batch(1, 50)[0]
source_df = pd.DataFrame({'feature_1': air[:-1], 'target': air[1:]})
source_df['category'] = ['1' for i in range(source_df.shape[0])]
hrv = pd.read_csv("RR_train.csv")
dataset_df = pd.DataFrame()
from deepar.dataset.time_series import TimeSeries
from deepar.model.lstm import DeepAR
from sklearn.preprocessing import MinMaxScaler
ts = TimeSeries(source_df, scaler=MinMaxScaler)
dp_model = DeepAR(ts, epochs=100)
dp_model.instantiate_and_fit()
%matplotlib inline
from numpy.random import normal
import tqdm
import pandas as pd
from matplotlib import pyplot as plt
import numpy as np
batch = ts.next_batch(1, 20)
def get_sample_prediction(sample, prediction_fn):