def train_classifier(when):
mldb.put(
"/v1/procedures/tng_classif", {
"type": "classifier.train",
"params": {
"trainingData": {
"select": "{* EXCLUDING (x)} as features, x as label",
"when": when,
"from": {
"id": "dataset1"
}
},
"configuration": {
"glz": {
"type": "glz",
"verbosity": 3,
"normalize": True,
"regularization": 'l2'
}
},
"algorithm": "glz",
"modelFileUrl": "file://tmp/MLDB-945.tng.cls"
}
})
mldb.post('/v1/procedures/tng_classif/runs')
def train_svd(when, output_index):
global dataset_index
dataset_index += 1
svd_procedure = "/v1/procedures/when_svd"
# svd procedure configuration
svd_config = {
'type': 'svd.train',
'params': {
"trainingData": {
"from": {
"id": "svd_example"
},
"when": when
},
"rowOutputDataset": {
"id": "when_svd_row_" + str(dataset_index),
'type': "embedding"
},
"columnOutputDataset": {
"id": "svd_embedding_" + str(output_index),
"type": "embedding"
}
}
}
mldb.put(svd_procedure, svd_config)
mldb.post(svd_procedure + '/runs')
result = mldb.get('/v1/query',
q="SELECT * FROM when_svd_row_" + str(dataset_index))
response = result.json()
return len(response[0]["columns"])
def train_tsne(when):
global dataset_index
dataset_index += 1
tsne_procedure = "/v1/procedures/when_tsne"
# t-sne procedure configuration
tsne_config = {
'type': 'tsne.train',
'params': {
"trainingData": {
"from": {
"id": "svd_example"
},
"when": when
},
"rowOutputDataset": {
"id": "tsne_embedding_" + str(dataset_index),
'type': "embedding"
}
}
}
mldb.put(tsne_procedure, tsne_config)
mldb.post(tsne_procedure + '/runs')
result = mldb.get('/v1/query',
q="SELECT * FROM tsne_embedding_" + str(dataset_index))
return len(result.json()[0]["columns"])
def test_most_frequent(self):
ds = mldb.create_dataset({
'id': 'most_freq_source',
'type': 'sparse.mutable'
})
row_num = 0
class Counter(object):
def __init__(self):
self.num = 0
def __next__(self):
self.num += 1
return self.num
vals = {
'a': 5,
'b': 4,
'c': 3,
'd': 2,
'e': 1,
'f': 1,
'g': 1,
'h': 1,
'i': 1,
'j': 1,
'k': 1,
'l': 1,
'm': 1,
}
c = Counter()
for k, count in vals.items():
for _ in range(count):
ds.record_row(next(c), [['col', k, 0]])
ds.commit()
mldb.post(
'/v1/procedures', {
'type': 'summary.statistics',
'params': {
'runOnCreation': True,
'inputData': "SELECT * FROM most_freq_source",
'outputDataset': {
'id': 'most_freq_output',
'type': 'sparse.mutable'
}
}
})
res = mldb.query("SELECT * FROM most_freq_output ORDER BY rowName()")
self.assertTableResultEquals(res, [[
"_rowName", "value.data_type", "value.most_frequent_items.a",
"value.most_frequent_items.b", "value.most_frequent_items.c",
"value.most_frequent_items.d", "value.most_frequent_items.h",
"value.most_frequent_items.i", "value.most_frequent_items.j",
"value.most_frequent_items.k", "value.most_frequent_items.l",
"value.most_frequent_items.m", "value.num_null", "value.num_unique"
], ["col", "categorical", 5, 4, 3, 2, 1, 1, 1, 1, 1, 1, 0, 13]])
def test_string_over_null(self):
"""
The primary test.
The issue column, which was always null for training, is now having
a string values for testing.
"""
ds = mldb.create_dataset({
'id': 'test_string_over_null_ds',
'type': 'sparse.mutable'
})
for idx in range(10):
ds.record_row(
'row{}'.format(idx),
[['line', idx, 0], ['label', 0, 0],
['feature', random.random() + 0.6, 0],
['noise', random.random(), 0], ['issue', 'STRING', 0]])
ds.commit()
mldb.post('/v1/procedures', {
'type': 'classifier.test',
'params' : {
'mode': 'boolean',
'testingData':
'SELECT score_it({features: {feature, noise, issue}})[score] AS score, ' \
'label FROM test_string_over_null_ds'
}
})
def test_record_null_row_name(self):
mldb.put('/v1/datasets/ds_null', {'type': 'sparse.mutable'})
with self.assertRaises(ResponseException): # noqa
mldb.post('/v1/datasets/ds_null/rows', {
'rowName': None,
'columns': [['colA', 1, 1]]
})
def test_set_return_0(self):
mldb.put(
"/v1/plugins/mldb2114", {
"type": "python",
"params": {
"source": {
"routes": """request.set_return("", 0)"""
}
}
})
with self.assertRaises(ResponseException) as e:
mldb.get('/v1/plugins/mldb2114/routes/foo')
self.assertEqual(e.exception.response.status_code, 500)
with self.assertRaises(ResponseException) as e:
mldb.post('/v1/plugins/mldb2114/routes/foo')
self.assertEqual(e.exception.response.status_code, 500)
with self.assertRaises(ResponseException) as e:
mldb.put('/v1/plugins/mldb2114/routes/foo')
self.assertEqual(e.exception.response.status_code, 500)
with self.assertRaises(ResponseException) as e:
mldb.delete('/v1/plugins/mldb2114/routes/foo')
self.assertEqual(e.exception.response.status_code, 500)
def test_no_set_return(self):
mldb.put(
"/v1/plugins/mldb2114", {
"type": "python",
"params": {
"source": {
"routes":
"""
from mldb import mldb
mldb.log('no return')
"""
}
}
})
msg = "Return value is required for route handlers but not set"
with self.assertRaisesRegex(ResponseException, msg) as e:
mldb.get('/v1/plugins/mldb2114/routes/foo')
self.assertEqual(e.exception.response.status_code, 500)
with self.assertRaisesRegex(ResponseException, msg) as e:
mldb.post('/v1/plugins/mldb2114/routes/foo')
self.assertEqual(e.exception.response.status_code, 500)
with self.assertRaisesRegex(ResponseException, msg) as e:
mldb.put('/v1/plugins/mldb2114/routes/foo')
self.assertEqual(e.exception.response.status_code, 500)
with self.assertRaisesRegex(ResponseException, msg) as e:
mldb.delete('/v1/plugins/mldb2114/routes/foo')
self.assertEqual(e.exception.response.status_code, 500)
def test_dottest_col_names(self):
ds = mldb.create_dataset({
'id': 'dotted_col_ds',
'type': 'sparse.mutable'
})
ds.record_row('row1', [['col.a', 1, 0]])
ds.commit()
mldb.post(
'/v1/procedures', {
'type': 'summary.statistics',
'params': {
'runOnCreation': True,
'inputData': 'SELECT * FROM dotted_col_ds',
'outputDataset': {
'id': 'output_dotted_col_ds',
'type': 'sparse.mutable'
}
}
})
res = mldb.query("SELECT * FROM output_dotted_col_ds")
self.assertTableResultEquals(res, [[
"_rowName", "value.1st_quartile", "value.3rd_quartile",
"value.data_type", "value.max", "value.avg", "value.median",
"value.min", "value.most_frequent_items.1", "value.num_null",
"value.num_unique", "value.stddev"
], ['"col.a"', 1, 1, "number", 1, 1, 1, 1, 1, 0, 1, "NaN"]])
def insert_with_ts(ts):
mldb.post('/v1/datasets/ds/rows', {
'rowName' : 'row1',
'columns' : [
['colA', 1, ts],
]
})
def test_import_missing_param(self):
msg = 'uriConnectionScheme is a required property'
with self.assertRaisesRegex(ResponseException, msg):
mldb.post('/v1/procedures', {
'type' : 'mongodb.import',
'params' : {
'collection' : 'users',
'outputDataset' : {
'id' : 'out',
'type' : 'sparse.mutable'
}
}
})
msg = 'collection is a required property and must not be empty'
with self.assertRaisesRegex(ResponseException, msg):
mldb.post('/v1/procedures', {
'type' : 'mongodb.import',
'params' : {
'uriConnectionScheme' : 'mongodb://localhost:27017/tutorial',
'outputDataset' : {
'id' : 'out',
'type' : 'sparse.mutable'
}
}
})
def test_rowHash(self):
mldb.post(
'/v1/procedures', {
'type': 'import.text',
'params': {
"dataFileUrl":
"https://raw.githubusercontent.com/datacratic/mldb-pytanic-plugin/master/titanic_train.csv",
'outputDataset': "titanic_hashed",
"where": "rowHash() % 3 = 0",
'runOnCreation': True,
}
})
mldb.post(
'/v1/procedures', {
'type': 'import.text',
'params': {
"dataFileUrl":
"https://raw.githubusercontent.com/datacratic/mldb-pytanic-plugin/master/titanic_train.csv",
'outputDataset': "titanic_no_hashed",
'runOnCreation': True,
}
})
self.assertTableResultEquals(
mldb.query("select count(*) from titanic_hashed"),
[["_rowName", "count(*)"], ["[]", 287]])
self.assertTableResultEquals(
mldb.query("select count(*) from titanic_no_hashed"),
[["_rowName", "count(*)"], ["[]", 891]])
def test_function_creation_bug(self):
mldb.post(
"/v1/procedures", {
"type": "import.text",
"params": {
"dataFileUrl": "http://public.mldb.ai/narrow_test.csv.gz",
"outputDataset": "narrow",
"runOnCreation": True
}
})
# it seems that the training fails to save the function but we proceed to testing
# where we try to use the function but then can't find it
# 1) we should not move to testing if function-creation fails
# we should report that function-creation failed
# 2) function creation should not fail for a dt on this dataset
mldb.put(
"/v1/procedures/train", {
"type": "classifier.experiment",
"params": {
"experimentName": "x",
"inputData":
"select {a} as features, b as label from narrow",
"algorithm": "dt",
"mode": "regression",
"configurationFile":
"./mldb/container_files/classifiers.json",
"modelFileUrlPattern":
"file://tmp/MLDB-1597-creation$runid.cls",
"runOnCreation": True
}
})
def train_svd(order_by, where, offset, limit):
svd_procedure = "/v1/procedures/order_svd"
# svd procedure configuration
svd_config = {
'type': 'svd.train',
'params': {
"trainingData": {
"from": {
"id": "svd_example"
},
"select": "x, y, z",
"orderBy": order_by,
"where": where,
"offset": offset,
"limit": limit
},
"rowOutputDataset": {
"id": "svd_row",
'type': "embedding"
},
"columnOutputDataset": {
"id": "svd_column",
"type": "embedding"
}
}
}
mldb.put(svd_procedure, svd_config)
mldb.post(svd_procedure + '/runs')
result = mldb.get('/v1/query', q="SELECT * FROM svd_row")
return len(result.json()[0]["columns"])
def test_do_not_run_on_creation(self):
mldb.put('/v1/procedures/do_not_run_on_creation', {
'type' : 'transform',
'params' : {
'skipEmptyRows' : False,
'runOnCreation' : False
}
})
msg = 'You need to define inputData'
with self.assertRaisesRegex(ResponseException, msg):
mldb.put('/v1/procedures/do_not_run_on_creation/runs/r1', {
'params' : {}
})
res = mldb.post('/v1/procedures', {
'type' : 'transform',
'params' : {
'skipEmptyRows' : False,
'runOnCreation' : False
}
}).json()
with self.assertRaisesRegex(ResponseException, msg):
mldb.post('/v1/procedures/{}/runs'.format(res['id']), {
'params' : {}
})
def test_join_with_and(self):
resp = mldb.query('select * from ds_train')
mldb.log(resp)
mldb.post(
"/v1/procedures", {
"type": "transform",
"params": {
"inputData": """
select *
from ds left join ds_stats on (ds.dow=ds_stats.dow and ds.a_int=ds_stats.a_int)
limit 10
""",
"outputDataset": {
"id": "ds_train2",
"type": "tabular",
"params": {
"unknownColumns": "add"
}
},
"runOnCreation": True
}
})
resp2 = mldb.query('select * from ds_train2')
mldb.log(resp2)
# equivalent join conditions should be returning the same dataset
# this is a very weak check because the columns and the row ordering
# of these two equivalent joins are currently very different
self.assertEqual(len(resp), len(resp2),
'expected response sizes to match')
def setUpClass(cls):
ds = mldb.create_dataset({'id' : 'ds', 'type' : 'sparse.mutable'})
ds.record_row('row1', [['label', 1, 12], ['feat1', 1, 0], ['feat2', 1, 0]])
ds.record_row('row2', [['label', 0, 12], ['feat1', 1, 0], ['feat2', 0, 0]])
ds.record_row('row3', [['label', 0, 12], ['feat1', 0, 0], ['feat2', 0, 0]])
ds.commit()
mldb.post('/v1/procedures', {
'type' : 'classifier.train',
'params' : {
'runOnCreation' : True,
"mode": "boolean",
'algorithm' : 'glz',
"configuration": {
"glz": {
"type": "glz",
"verbosity": 3,
"normalize": False,
"regularization": 'l2'
}
},
'trainingData' : """
SELECT {* EXCLUDING(label)} AS features, label
FROM ds
""",
"modelFileUrl":
"file://build/x86_64/tmp/fmlhTODO.cls",
}
})
def test_c_over_a_or_b(self):
"""
This is an alternate test of unknown values. The model was built with
column a_or_b having always a value of either a or b. Here, we test
with a dataset having always the "never seen value" of c for that
column.
"""
ds = mldb.create_dataset({
'id': 'test_c_over_a_or_b_ds',
'type': 'sparse.mutable'
})
for idx in range(10):
ds.record_row('row{}'.format(idx),
[['line', idx, 0], ['label', 0, 0],
['feature', random.random() + 0.6, 0],
['noise', random.random(), 0], ['a_or_b', 'c', 0]])
ds.commit()
mldb.post('/v1/procedures', {
'type': 'classifier.test',
'params' : {
'mode': 'boolean',
'testingData':
'SELECT score_it({features: {feature, noise, a_or_b}})[score] AS score, ' \
'label FROM test_c_over_a_or_b_ds'
}
})
def test_too_many_requested_rows(self):
# too many requested rows without sampling
sampled_dataset_conf = {
"type": "sampled",
"params": {
"dataset": {
"id": "toy"
},
"rows": 25000,
"withReplacement": False
}
}
with self.assertRaises(ResponseException) as re:
mldb.put("/v1/datasets/patate", sampled_dataset_conf)
self.assertEqual(re.exception.response.status_code, 400)
sampled_dataset_conf["params"]["withReplacement"] = True
mldb.put("/v1/datasets/patate", sampled_dataset_conf)
# try to insert and make sure we get an exception
with self.assertRaises(ResponseException) as re:
mldb.post("/v1/datasets/patate/rows", {
"rowName": "patato",
"columns": [["a", "b", 0]]
})
self.assertEqual(re.exception.response.status_code, 400)
def setUpClass(cls):
mldb.post(
'/v1/procedures', {
'type': 'import.text',
'params': {
'dataFileUrl':
'http://public.mldb.ai/datasets/enron.csv.gz',
'outputDataset': 'enron_data',
'named': "'enron_' + dataset + '_mail_' + index",
'where': 'dataset = 1'
}
})
mldb.put(
'/v1/functions/bow', {
'type': 'sql.expression',
'params': {
'expression':
"""
tokenize(msg, {splitChars: ' :.-!?''"()[],', quoteChar: ''}) as bow
"""
}
})
mldb.post(
'/v1/procedures', {
'type': 'transform',
'params': {
'inputData': """
select bow({msg})[bow] as *, label = 'spam' as message_is_spam
from enron_data
""",
'outputDataset': 'enron_features'
}
})
def test_delete(self):
# this test depends on put and post
url = '/v1/datasets/ds'
mldb.put(url, {
'type' : 'sparse.mutable'
})
mldb.post(url + '/commit')
mldb.delete(url)
def test_offset_not_accepted(self):
with self.assertRaises(ResponseException):
mldb.post(
"/v1/procedures", {
"type": "classifier.experiment",
"params": {
'runOnCreation': True,
"experimentName": "test_no_test_data",
"inputData": """
SELECT {* EXCLUDING(label)} AS features, label
FROM toy OFFSET 1""",
"modelFileUrlPattern":
"file://build/x86_64/tmp/test_no_test_data_$runid.cls",
"algorithm": "glz",
"equalizationFactor": 0.5,
"mode": "boolean",
"configuration": {
"glz": {
"type": "glz",
"verbosity": 3,
"normalize": False,
"regularization": 'l2'
}
},
"outputAccuracyDataset": False
}
})
with self.assertRaises(ResponseException):
mldb.post(
"/v1/procedures", {
"type": "classifier.experiment",
"params": {
'runOnCreation': True,
"experimentName": "test_no_test_data",
"inputData": """
SELECT {* EXCLUDING(label)} AS features, label
FROM toy""",
"testingDataOverride": """
SELECT {* EXCLUDING(label)} AS features, label
FROM toy OFFSET 1""",
"modelFileUrlPattern":
"file://build/x86_64/tmp/test_no_test_data_$runid.cls",
"algorithm": "glz",
"equalizationFactor": 0.5,
"mode": "boolean",
"configuration": {
"glz": {
"type": "glz",
"verbosity": 3,
"normalize": False,
"regularization": 'l2'
}
},
"outputAccuracyDataset": False
}
})
def test_post_empty_row_name(self):
mldb.put('/v1/datasets/empty_row_name', {'type': 'sparse.mutable'})
mldb.post('/v1/datasets/empty_row_name/rows', {
'rowName': '',
'columns': [['col', 'val', 0]]
})
mldb.post('/v1/datasets/empty_row_name/commit')
res = mldb.query("SELECT * FROM empty_row_name")
self.assertTableResultEquals(res, [['_rowName', 'col'], ['""', "val"]])
def setUpClass(cls):
mldb.post(
"/v1/functions", {
"id": "explain",
"type": "classifier.explain",
"params": {
"modelFileUrl":
"file://mldb/testing/fixtures/mnist_glz_categorical.cls"
}
})
def test_post_row_name_none(self):
mldb.put('/v1/datasets/ds2', {'type': 'sparse.mutable'})
with self.assertRaises(ResponseException): # noqa
mldb.post(
'/v1/datasets/ds2/rows',
{
'rowName': None, # should not work
'columns': [['colA', 1, 1]]
})
def test_post_no_row_name(self):
mldb.put('/v1/datasets/ds3', {'type': 'sparse.mutable'})
with self.assertRaises(ResponseException): # noqa
mldb.post(
'/v1/datasets/ds3/rows',
{
# rowName missing -> should not work
'columns': [['colA', 1, 1]]
})
def test_post(self):
res = mldb.post("/v1/datasets", {
'type' : 'sparse.mutable'
})
id_ = res.json()['id']
url = '/v1/datasets/{}'.format(id_)
mldb.post(url + '/rows', {
'rowName' : 'row1',
'columns' : [['colA', 1, 0]]
})
mldb.post(url + '/commit')
def test_query(self):
url = '/v1/datasets/ds'
mldb.put(url, {
'type' : 'sparse.mutable'
})
mldb.post(url + '/rows', {
'rowName' : 'row1',
'columns' : [['colA', 1, 0]]
})
mldb.post(url + '/commit')
res = mldb.query('SELECT * FROM ds')
self.assertEqual(res, [['_rowName', 'colA'], ['row1', 1]])
def run_proc(input_data):
mldb.post(
'/v1/procedures', {
'type': 'summary.statistics',
'params': {
'runOnCreation': True,
'inputData': input_data,
'outputDataset': {
'id': 'error',
'type': 'sparse.mutable'
}
}
})
def do_it(self, ds_type):
mldb.post(
"/v1/procedures", {
"type": "transform",
"params": {
"inputData": self.__class__.query,
"outputDataset": {
"id": ds_type,
"type": ds_type
},
'runOnCreation': True
}
})