Spaces:
Runtime error
Runtime error
File size: 11,073 Bytes
4a51346 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 |
import pytest
import logging
import hypothesis.strategies as st
from typing import Set, cast, Union, DefaultDict
from dataclasses import dataclass
from chromadb.api.types import ID, Include, IDs
import chromadb.errors as errors
from chromadb.api import API
from chromadb.api.models.Collection import Collection
import chromadb.test.property.strategies as strategies
from hypothesis.stateful import (
Bundle,
RuleBasedStateMachine,
MultipleResults,
rule,
initialize,
precondition,
consumes,
run_state_machine_as_test,
multiple,
invariant,
)
from collections import defaultdict
import chromadb.test.property.invariants as invariants
import numpy as np
traces: DefaultDict[str, int] = defaultdict(lambda: 0)
def trace(key: str) -> None:
global traces
traces[key] += 1
def print_traces() -> None:
global traces
for key, value in traces.items():
print(f"{key}: {value}")
dtype_shared_st: st.SearchStrategy[
Union[np.float16, np.float32, np.float64]
] = st.shared(st.sampled_from(strategies.float_types), key="dtype")
dimension_shared_st: st.SearchStrategy[int] = st.shared(
st.integers(min_value=2, max_value=2048), key="dimension"
)
@dataclass
class EmbeddingStateMachineStates:
initialize = "initialize"
add_embeddings = "add_embeddings"
delete_by_ids = "delete_by_ids"
update_embeddings = "update_embeddings"
upsert_embeddings = "upsert_embeddings"
collection_st = st.shared(strategies.collections(with_hnsw_params=True), key="coll")
class EmbeddingStateMachine(RuleBasedStateMachine):
collection: Collection
embedding_ids: Bundle[ID] = Bundle("embedding_ids")
def __init__(self, api: API):
super().__init__()
self.api = api
self._rules_strategy = strategies.DeterministicRuleStrategy(self) # type: ignore
@initialize(collection=collection_st) # type: ignore
def initialize(self, collection: strategies.Collection):
self.api.reset()
self.collection = self.api.create_collection(
name=collection.name,
metadata=collection.metadata,
embedding_function=collection.embedding_function,
)
self.embedding_function = collection.embedding_function
trace("init")
self.on_state_change(EmbeddingStateMachineStates.initialize)
self.record_set_state = strategies.StateMachineRecordSet(
ids=[], metadatas=[], documents=[], embeddings=[]
)
@rule(target=embedding_ids, record_set=strategies.recordsets(collection_st))
def add_embeddings(self, record_set: strategies.RecordSet) -> MultipleResults[ID]:
trace("add_embeddings")
self.on_state_change(EmbeddingStateMachineStates.add_embeddings)
normalized_record_set: strategies.NormalizedRecordSet = invariants.wrap_all(
record_set
)
if len(normalized_record_set["ids"]) > 0:
trace("add_more_embeddings")
if set(normalized_record_set["ids"]).intersection(
set(self.record_set_state["ids"])
):
with pytest.raises(errors.IDAlreadyExistsError):
self.collection.add(**normalized_record_set)
return multiple()
else:
self.collection.add(**normalized_record_set)
self._upsert_embeddings(cast(strategies.RecordSet, normalized_record_set))
return multiple(*normalized_record_set["ids"])
@precondition(lambda self: len(self.record_set_state["ids"]) > 20)
@rule(ids=st.lists(consumes(embedding_ids), min_size=1, max_size=20))
def delete_by_ids(self, ids: IDs) -> None:
trace("remove embeddings")
self.on_state_change(EmbeddingStateMachineStates.delete_by_ids)
indices_to_remove = [self.record_set_state["ids"].index(id) for id in ids]
self.collection.delete(ids=ids)
self._remove_embeddings(set(indices_to_remove))
# Removing the precondition causes the tests to frequently fail as "unsatisfiable"
# Using a value < 5 causes retries and lowers the number of valid samples
@precondition(lambda self: len(self.record_set_state["ids"]) >= 5)
@rule(
record_set=strategies.recordsets(
collection_strategy=collection_st,
id_strategy=embedding_ids,
min_size=1,
max_size=5,
)
)
def update_embeddings(self, record_set: strategies.RecordSet) -> None:
trace("update embeddings")
self.on_state_change(EmbeddingStateMachineStates.update_embeddings)
self.collection.update(**record_set)
self._upsert_embeddings(record_set)
# Using a value < 3 causes more retries and lowers the number of valid samples
@precondition(lambda self: len(self.record_set_state["ids"]) >= 3)
@rule(
record_set=strategies.recordsets(
collection_strategy=collection_st,
id_strategy=st.one_of(embedding_ids, strategies.safe_text),
min_size=1,
max_size=5,
)
)
def upsert_embeddings(self, record_set: strategies.RecordSet) -> None:
trace("upsert embeddings")
self.on_state_change(EmbeddingStateMachineStates.upsert_embeddings)
self.collection.upsert(**record_set)
self._upsert_embeddings(record_set)
@invariant()
def count(self) -> None:
invariants.count(
self.collection, cast(strategies.RecordSet, self.record_set_state)
)
@invariant()
def no_duplicates(self) -> None:
invariants.no_duplicates(self.collection)
@invariant()
def ann_accuracy(self) -> None:
invariants.ann_accuracy(
collection=self.collection,
record_set=cast(strategies.RecordSet, self.record_set_state),
min_recall=0.95,
embedding_function=self.embedding_function,
)
def _upsert_embeddings(self, record_set: strategies.RecordSet) -> None:
normalized_record_set: strategies.NormalizedRecordSet = invariants.wrap_all(
record_set
)
for idx, id in enumerate(normalized_record_set["ids"]):
# Update path
if id in self.record_set_state["ids"]:
target_idx = self.record_set_state["ids"].index(id)
if normalized_record_set["embeddings"] is not None:
self.record_set_state["embeddings"][
target_idx
] = normalized_record_set["embeddings"][idx]
else:
assert normalized_record_set["documents"] is not None
assert self.embedding_function is not None
self.record_set_state["embeddings"][
target_idx
] = self.embedding_function(
[normalized_record_set["documents"][idx]]
)[
0
]
if normalized_record_set["metadatas"] is not None:
self.record_set_state["metadatas"][
target_idx
] = normalized_record_set["metadatas"][idx]
if normalized_record_set["documents"] is not None:
self.record_set_state["documents"][
target_idx
] = normalized_record_set["documents"][idx]
else:
# Add path
self.record_set_state["ids"].append(id)
if normalized_record_set["embeddings"] is not None:
self.record_set_state["embeddings"].append(
normalized_record_set["embeddings"][idx]
)
else:
assert self.embedding_function is not None
assert normalized_record_set["documents"] is not None
self.record_set_state["embeddings"].append(
self.embedding_function(
[normalized_record_set["documents"][idx]]
)[0]
)
if normalized_record_set["metadatas"] is not None:
self.record_set_state["metadatas"].append(
normalized_record_set["metadatas"][idx]
)
else:
self.record_set_state["metadatas"].append(None)
if normalized_record_set["documents"] is not None:
self.record_set_state["documents"].append(
normalized_record_set["documents"][idx]
)
else:
self.record_set_state["documents"].append(None)
def _remove_embeddings(self, indices_to_remove: Set[int]) -> None:
indices_list = list(indices_to_remove)
indices_list.sort(reverse=True)
for i in indices_list:
del self.record_set_state["ids"][i]
del self.record_set_state["embeddings"][i]
del self.record_set_state["metadatas"][i]
del self.record_set_state["documents"][i]
def on_state_change(self, new_state: str) -> None:
pass
def test_embeddings_state(caplog: pytest.LogCaptureFixture, api: API) -> None:
caplog.set_level(logging.ERROR)
run_state_machine_as_test(lambda: EmbeddingStateMachine(api)) # type: ignore
print_traces()
def test_multi_add(api: API) -> None:
api.reset()
coll = api.create_collection(name="foo")
coll.add(ids=["a"], embeddings=[[0.0]])
assert coll.count() == 1
with pytest.raises(errors.IDAlreadyExistsError):
coll.add(ids=["a"], embeddings=[[0.0]])
assert coll.count() == 1
results = coll.get()
assert results["ids"] == ["a"]
coll.delete(ids=["a"])
assert coll.count() == 0
def test_dup_add(api: API) -> None:
api.reset()
coll = api.create_collection(name="foo")
with pytest.raises(errors.DuplicateIDError):
coll.add(ids=["a", "a"], embeddings=[[0.0], [1.1]])
with pytest.raises(errors.DuplicateIDError):
coll.upsert(ids=["a", "a"], embeddings=[[0.0], [1.1]])
def test_query_without_add(api: API) -> None:
api.reset()
coll = api.create_collection(name="foo")
fields: Include = ["documents", "metadatas", "embeddings", "distances"]
N = np.random.randint(1, 2000)
K = np.random.randint(1, 100)
results = coll.query(
query_embeddings=np.random.random((N, K)).tolist(), include=fields
)
for field in fields:
field_results = results[field]
assert field_results is not None
assert all([len(result) == 0 for result in field_results])
# TODO: Use SQL escaping correctly internally
@pytest.mark.xfail(reason="We don't properly escape SQL internally, causing problems")
def test_escape_chars_in_ids(api: API) -> None:
api.reset()
id = "\x1f"
coll = api.create_collection(name="foo")
coll.add(ids=[id], embeddings=[[0.0]])
assert coll.count() == 1
coll.delete(ids=[id])
assert coll.count() == 0
|