-
Notifications
You must be signed in to change notification settings - Fork 160
/
demo.py
134 lines (112 loc) · 4.43 KB
/
demo.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
import os
import torch
os.environ['OMP_NUM_THREADS'] = '4'
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import openai
openai.api_key = os.environ['OPENAI_API_KEY']
openai.api_base = os.environ.get("OPENAI_API_BASE", "https://api.openai.com/v1")
from data.serialize import SerializerSettings
from models.utils import grid_iter
from models.promptcast import get_promptcast_predictions_data
from models.darts import get_arima_predictions_data
from models.llmtime import get_llmtime_predictions_data
from data.small_context import get_datasets
from models.validation_likelihood_tuning import get_autotuned_predictions_data
def plot_preds(train, test, pred_dict, model_name, show_samples=False):
pred = pred_dict['median']
pred = pd.Series(pred, index=test.index)
plt.figure(figsize=(8, 6), dpi=100)
plt.plot(train)
plt.plot(test, label='Truth', color='black')
plt.plot(pred, label=model_name, color='purple')
# shade 90% confidence interval
samples = pred_dict['samples']
lower = np.quantile(samples, 0.05, axis=0)
upper = np.quantile(samples, 0.95, axis=0)
plt.fill_between(pred.index, lower, upper, alpha=0.3, color='purple')
if show_samples:
samples = pred_dict['samples']
# convert df to numpy array
samples = samples.values if isinstance(samples, pd.DataFrame) else samples
for i in range(min(10, samples.shape[0])):
plt.plot(pred.index, samples[i], color='purple', alpha=0.3, linewidth=1)
plt.legend(loc='upper left')
if 'NLL/D' in pred_dict:
nll = pred_dict['NLL/D']
if nll is not None:
plt.text(0.03, 0.85, f'NLL/D: {nll:.2f}', transform=plt.gca().transAxes, bbox=dict(facecolor='white', alpha=0.5))
plt.show()
print(torch.cuda.max_memory_allocated())
print()
gpt4_hypers = dict(
alpha=0.3,
basic=True,
temp=1.0,
top_p=0.8,
settings=SerializerSettings(base=10, prec=3, signed=True, time_sep=', ', bit_sep='', minus_sign='-')
)
mistral_api_hypers = dict(
alpha=0.3,
basic=True,
temp=1.0,
top_p=0.8,
settings=SerializerSettings(base=10, prec=3, signed=True, time_sep=', ', bit_sep='', minus_sign='-')
)
gpt3_hypers = dict(
temp=0.7,
alpha=0.95,
beta=0.3,
basic=False,
settings=SerializerSettings(base=10, prec=3, signed=True, half_bin_correction=True)
)
llma2_hypers = dict(
temp=0.7,
alpha=0.95,
beta=0.3,
basic=False,
settings=SerializerSettings(base=10, prec=3, signed=True, half_bin_correction=True)
)
promptcast_hypers = dict(
temp=0.7,
settings=SerializerSettings(base=10, prec=0, signed=True,
time_sep=', ',
bit_sep='',
plus_sign='',
minus_sign='-',
half_bin_correction=False,
decimal_point='')
)
arima_hypers = dict(p=[12,30], d=[1,2], q=[0])
model_hypers = {
'LLMTime GPT-3.5': {'model': 'gpt-3.5-turbo-instruct', **gpt3_hypers},
'LLMTime GPT-4': {'model': 'gpt-4', **gpt4_hypers},
'LLMTime GPT-3': {'model': 'text-davinci-003', **gpt3_hypers},
'PromptCast GPT-3': {'model': 'text-davinci-003', **promptcast_hypers},
'LLMA2': {'model': 'llama-7b', **llma2_hypers},
'mistral': {'model': 'mistral', **llma2_hypers},
'mistral-api-tiny': {'model': 'mistral-api-tiny', **mistral_api_hypers},
'mistral-api-small': {'model': 'mistral-api-tiny', **mistral_api_hypers},
'mistral-api-medium': {'model': 'mistral-api-tiny', **mistral_api_hypers},
'ARIMA': arima_hypers,
}
model_predict_fns = {
#'LLMA2': get_llmtime_predictions_data,
#'mistral': get_llmtime_predictions_data,
#'LLMTime GPT-4': get_llmtime_predictions_data,
'mistral-api-tiny': get_llmtime_predictions_data
}
model_names = list(model_predict_fns.keys())
datasets = get_datasets()
ds_name = 'AirPassengersDataset'
data = datasets[ds_name]
train, test = data # or change to your own data
out = {}
for model in model_names: # GPT-4 takes a about a minute to run
model_hypers[model].update({'dataset_name': ds_name}) # for promptcast
hypers = list(grid_iter(model_hypers[model]))
num_samples = 10
pred_dict = get_autotuned_predictions_data(train, test, hypers, num_samples, model_predict_fns[model], verbose=False, parallel=False)
out[model] = pred_dict
plot_preds(train, test, pred_dict, model, show_samples=True)