class NativeDataFrame(BasePredictor):
def __init__(self, dataset):
super().__init__(dataset)
self.df = pd.read_csv(f"{self.dataset.name}_test.csv")
self.predictor = Predictor(name=self.dataset.name)
def predict(self, row_number=1):
return self.predictor.predict(self.df[:row_number])
class NativeClickhouse(BasePredictor):
host = CONFIG['database']['host']
port = CONFIG['database']['port']
user = CONFIG['database']['user']
password = CONFIG['database']['password']
def __init__(self, dataset):
super().__init__(dataset)
self.predictor = Predictor(name=self.dataset.name)
self.query_template = f"SELECT * FROM test_data.{self.dataset.name} LIMIT %s"
def predict(self, row_number=1):
_query = self.query_template % row_number
return self.predictor.predict(when_data=ClickhouseDS(
_query, host=self.host, user=self.user, password=self.password))
def test_regressor():
"""
Sanity check. MindsDB point predictions should be within range
of predicted bounds by the inductive conformal predictor.
"""
def _df_from_x(x, columns=None):
x = pd.DataFrame(x)
if columns is None:
x.columns = 'c' + pd.Series([i for i in range(len(x.columns))
]).astype(str)
else:
x.columns = columns
return x
def _df_from_xy(x, y, target):
x = _df_from_x(x)
x[target] = pd.DataFrame(y)
return x
X, y = load_boston(return_X_y=True)
X_train, X_test, Y_train, Y_test = train_test_split(X,
y,
test_size=0.1,
random_state=5)
target = 'medv'
x_tr = _df_from_xy(X_train, Y_train, target)
p = Predictor("ConformalTest")
p.learn(from_data=x_tr, to_predict=target)
x_te = _df_from_xy(X_test, Y_test, target)
r = p.predict(when_data=x_te)
r = [x.explanation[target] for x in r]
for x in r:
assert x['confidence_interval'][0] <= x['predicted_value'] <= x[
'confidence_interval'][1]
import pandas as pd
from mindsdb_native import Predictor
mdb = Predictor(name='description_predictor')
mdb.learn(from_data=pd.read_csv('processed_data/train.csv'),
to_predict='description')
predictions = mdb.predict('processed_data/test.csv')
for pred in predictions:
print(pred['description'])
def test_database_history(self):
return
from mindsdb_datasources import ClickhouseDS
TEMP_DB = 'test_database_history_' + random_string()
TEMP_TABLE = 'tmp_test_database_history_' + random_string()
params = {'user': self.USER, 'password': self.PASSWORD}
clickhouse_url = f'http://{self.HOST}:{self.PORT}'
values = []
for i in range(200):
values.append([str(i % 4), i, i * 2])
queries = [
f'CREATE DATABASE IF NOT EXISTS {TEMP_DB}',
f'DROP TABLE IF EXISTS {TEMP_DB}.{TEMP_TABLE}',
f'''
CREATE TABLE {TEMP_DB}.{TEMP_TABLE}(
col1 String
,col2 Int64
,col3 Int64
) ENGINE = MergeTree()
ORDER BY col2
PARTITION BY col1
''',
]
gc.collect()
for value in values:
value_ins_str = str(value).replace('[','').replace(']','')
queries.append(f"INSERT INTO {TEMP_DB}.{TEMP_TABLE} VALUES ({value_ins_str})")
for q in queries:
r = requests.post(clickhouse_url, data=q, params=params)
assert r.status_code == 200, r.text
clickhouse_ds = ClickhouseDS(
f'SELECT * FROM {TEMP_DB}.{TEMP_TABLE}',
host=self.HOST,
port=self.PORT,
user=self.USER,
password=self.PASSWORD
)
temp_predictor = Predictor(name='query_history_based_ts_predictor')
temp_predictor.learn(
to_predict='col3',
from_data=clickhouse_ds,
stop_training_in_x_seconds=5,
timeseries_settings={
'order_by': ['col2']
,'window': 6
,'group_by': ['col1']
}
)
del temp_predictor
ts_predictor = Predictor(name='query_history_based_ts_predictor')
predictions = ts_predictor.predict(
when_data={
'col2': 800
,'col1': '2'
},
advanced_args={
'use_database_history': True
}
)
assert predictions[0]['col3'] is not None
r = requests.post(
clickhouse_url,
data=f'DROP DATABASE {TEMP_DB}',
params=params
)
assert r.status_code == 200, 'failed to drop temporary database "{}"'.format(TEMP_DB)
