ArizeDatasetsClient to v8’s ArizeClient.experiments.
from arize.experimental.datasets import ArizeDatasetsClient
# v7 api_key parameter took developer key values
client = ArizeDatasetsClient(
api_key="your-developer-key" # Developer key (deprecated)
)
from arize import ArizeClient
# v8 api_key parameter takes standard API keys
client = ArizeClient(
api_key="your-api-key" # Standard API key
)
run_experiment()
Therun_experiment() method migrates from client.run_experiment() to client.experiments.run().
Parameter Reference
| Parameter | v7 | v8 | Changes |
|---|---|---|---|
space_id | Required | ❌ Removed | Not needed in v8 |
experiment_name | Required | Required | Renamed to name |
name | N/A | ✅ Required | Renamed from experiment_name |
task | Required | Required | — |
dataset_df | Optional | ❌ Removed | Dataset downloaded automatically in v8, use dataset_id |
dataset_id | Optional | Required | Now required; no longer accepts dataset_name |
dataset_name | Optional | ❌ Removed | Use dataset_id instead |
evaluators | Optional | Optional | — |
dry_run | Optional | Optional | — |
dry_run_count | N/A | ✅ Optional | Number of examples to use in dry run (default 10) |
concurrency | Optional | Optional | — |
set_global_tracer_provider | Optional | Optional | — |
exit_on_error | Optional | Optional | — |
Side-by-Side Comparison
from arize.experimental.datasets import ArizeDatasetsClient
# Client initialization
client = ArizeDatasetsClient(api_key="your-developer-key")
# Run experiment
experiment_id, results_df = client.run_experiment(
space_id="your-space-id",
experiment_name="my-experiment",
task=my_task,
dataset_id="dataset-123", # or dataset_name="my-dataset"
evaluators=my_evaluators,
dry_run=False,
concurrency=3,
set_global_tracer_provider=False,
exit_on_error=False
)
from arize import ArizeClient
# Client initialization
client = ArizeClient(api_key="your-api-key")
# Run experiment
experiment, results_df = client.experiments.run(
name="my-experiment", # Renamed from experiment_name
dataset_id="dataset-123", # Required, no dataset_name option
task=my_task,
evaluators=my_evaluators,
dry_run=False,
dry_run_count=10, # New parameter
concurrency=3,
set_global_tracer_provider=False,
exit_on_error=False
# space_id removed
# dataset_df removed (downloaded automatically)
)
experiment_id = experiment.id if experiment else None
log_experiment()
Thelog_experiment() method migrates from client.log_experiment() to client.experiments.create().
Parameter Reference
| Parameter | v7 | v8 | Changes |
|---|---|---|---|
space_id | Required | ❌ Removed | Not needed in v8 |
experiment_name | Required | Required | Renamed to name |
name | N/A | ✅ Required | Renamed from experiment_name |
experiment_df | Required | Required | Renamed to experiment_runs |
experiment_runs | N/A | ✅ Required | Renamed from experiment_df; accepts DataFrame or list of dicts |
task_columns | Required | Required | Renamed to task_fields |
task_fields | N/A | ✅ Required | Renamed from task_columns |
evaluator_columns | Optional | Optional | — |
dataset_id | Optional | Required | Now required; no longer accepts dataset_name |
dataset_name | Optional | ❌ Removed | Use dataset_id instead |
force_http | N/A | ✅ Optional | Force HTTP upload instead of gRPC (default False) |
Side-by-Side Comparison
from arize.experimental.datasets import ArizeDatasetsClient
from arize.experimental.datasets.experiments.types import (
ExperimentTaskResultColumnNames,
EvaluationResultColumnNames
)
# Client initialization
client = ArizeDatasetsClient(api_key="your-developer-key")
# Log experiment
experiment_id = client.log_experiment(
space_id="your-space-id",
experiment_name="my-experiment",
experiment_df=results_df,
task_columns=ExperimentTaskResultColumnNames(
example_id="example_id",
result="output"
),
evaluator_columns={
"accuracy": EvaluationResultColumnNames(
score="accuracy_score"
)
},
dataset_id="dataset-123" # or dataset_name="my-dataset"
)
from arize import ArizeClient
from arize.experiments import ExperimentTaskFieldNames, EvaluationResultFieldNames
# Client initialization
client = ArizeClient(api_key="your-api-key")
# Create experiment
experiment = client.experiments.create(
name="my-experiment", # Renamed from experiment_name
dataset_id="dataset-123", # Required, no dataset_name option
experiment_runs=results_df, # Renamed from experiment_df
task_fields=ExperimentTaskFieldNames( # Renamed from task_columns
example_id="example_id",
output="output"
),
evaluator_columns={
"accuracy": EvaluationResultFieldNames(
score="accuracy_score"
)
},
force_http=False # New parameter
# space_id removed
)
experiment_id = experiment.id
get_experiment()
Theget_experiment() method has a different behavior in v8. In v7, client.get_experiment() returned the experiment data including all runs. In v8, client.experiments.get() returns only the experiment metadata, while client.experiments.list_runs() retrieves the actual experiment runs.
Parameter Reference
For experiment metadata (v8’sexperiments.get()):
| Parameter | v7 | v8 | Changes |
|---|---|---|---|
space_id | Required | ❌ Removed | Not needed in v8 |
experiment_id | Optional | Required | Now required and simplified |
experiment_name | Optional | ❌ Removed | Use experiment_id instead |
dataset_name | Optional | ❌ Removed | Use experiment_id instead |
experiments.list_runs()):
| Parameter | v7 | v8 | Changes |
|---|---|---|---|
experiment_id | Optional | Required | — |
limit | N/A | ✅ Optional | Maximum number of runs per page (default 100); ignored if all=True |
all | N/A | ✅ Optional | When True, retrieves all runs via Flight (bypasses pagination). When False (default), uses REST with pagination |
Side-by-Side Comparison
from arize.experimental.datasets import ArizeDatasetsClient
# Client initialization
client = ArizeDatasetsClient(api_key="your-developer-key")
# Get experiment data (includes runs)
experiment_df = client.get_experiment(
space_id="your-space-id",
experiment_id="exp-123"
)
# Returns: pandas DataFrame with experiment runs
# Or get by name
experiment_df = client.get_experiment(
space_id="your-space-id",
experiment_name="my-experiment",
dataset_name="my-dataset"
)
from arize import ArizeClient
# Client initialization
client = ArizeClient(api_key="your-api-key")
# Step 1: Get experiment metadata
experiment = client.experiments.get(experiment_id="exp-123")
# Returns: Experiment object with metadata
# Does NOT include runs (experiment data)
# Step 2: Get experiment runs (actual data)
# Option A: Get all runs at once (uses Flight, more efficient for large experiments)
runs_response = client.experiments.list_runs(
experiment_id="exp-123",
all=True # Retrieves all runs via Flight (bypasses pagination)
)
runs = runs_response.to_df()
# Option B: Use pagination (uses REST, better for browsing/previewing)
runs_response = client.experiments.list_runs(
experiment_id="exp-123",
limit=100, # Get 100 runs per page
all=False # Use REST with pagination (default)
)
runs = runs_response.to_df()
# Use response.pagination.cursor for next page if needed
delete_experiment()
Thedelete_experiment() method migrates from client.delete_experiment() to client.experiments.delete().
Parameter Reference
| Parameter | v7 | v8 | Changes |
|---|---|---|---|
space_id | Required | ❌ Removed | Not needed in v8 |
experiment_id | Optional | Required | Now required; no longer accepts experiment_name + dataset_name |
experiment_name | Optional | ❌ Removed | Use experiment_id instead |
dataset_name | Optional | ❌ Removed | Use experiment_id instead |
Side-by-Side Comparison
from arize.experimental.datasets import ArizeDatasetsClient
# Client initialization
client = ArizeDatasetsClient(api_key="your-developer-key")
# Delete experiment by ID
success, deleted_id = client.delete_experiment(
space_id="your-space-id",
experiment_id="exp-123"
)
# Or delete by name
success, deleted_id = client.delete_experiment(
space_id="your-space-id",
experiment_name="my-experiment",
dataset_name="my-dataset"
)
from arize import ArizeClient
# Client initialization
client = ArizeClient(api_key="your-api-key")
# Delete experiment by ID
client.experiments.delete(experiment_id="exp-123")
# Returns None on success