Import your data

How to create Phoenix inferences and schemas for common data formats

This guide shows you how to define Phoenix inferences using your own data.

• For a conceptual overview of the Phoenix API, including a high-level introduction to the notion of inferences and schemas, see here.

• For a comprehensive description of phoenix.Dataset and phoenix.Schema, see the API reference.

Once you have a pandas dataframe df containing your data and a schema object describing the format of your dataframe, you can define your Phoenix dataset either by running

ds = px.Inferences(df, schema)

or by optionally providing a name for your dataset that will appear in the UI:

ds = px.Inferences(df, schema, name="training")

As you can see, instantiating your dataset is the easy part. Before you run the code above, you must first wrangle your data into a pandas dataframe and then create a Phoenix schema to describe the format of your dataframe. The rest of this guide shows you how to match your schema to your dataframe with concrete examples.

Predictions and Actuals

Let’s first see how to define a schema with predictions and actuals (Phoenix’s nomenclature for ground truth). The example dataframe below contains inference data from a binary classification model trained to predict whether a user will click on an advertisement. The timestamps are datetime.datetime objects that represent the time at which each inference was made in production.

Dataframe

timestamp	prediction_score	prediction	target
2023-03-01 02:02:19	0.91	click	click
2023-02-17 23:45:48	0.37	no_click	no_click
2023-01-30 15:30:03	0.54	click	no_click
2023-02-03 19:56:09	0.74	click	click
2023-02-24 04:23:43	0.37	no_click	click

Schema

schema = px.Schema(
    timestamp_column_name="timestamp",
    prediction_score_column_name="prediction_score",
    prediction_label_column_name="prediction",
    actual_label_column_name="target",
)

This schema defines predicted and actual labels and scores, but you can run Phoenix with any subset of those fields, e.g., with only predicted labels.

Features and Tags

Phoenix accepts not only predictions and ground truth but also input features of your model and tags that describe your data. In the example below, features such as FICO score and merchant ID are used to predict whether a credit card transaction is legitimate or fraudulent. In contrast, tags such as age and gender are not model inputs, but are used to filter your data and analyze meaningful cohorts in the app.

Dataframe

fico_score	merchant_id	loan_amount	annual_income	home_ownership	num_credit_lines	inquests_in_last_6_months	months_since_last_delinquency	age	gender	predicted	target
578	Scammeds	4300	62966	RENT	110	0	0	25	male	not_fraud	fraud
507	Schiller Ltd	21000	52335	RENT	129	0	23	78	female	not_fraud	not_fraud
656	Kirlin and Sons	18000	94995	MORTGAGE	31	0	0	54	female	uncertain	uncertain
414	Scammeds	18000	32034	LEASE	81	2	0	34	male	fraud	not_fraud
512	Champlin and Sons	20000	46005	OWN	148	1	0	49	male	uncertain	uncertain

Schema

schema = px.Schema(
    prediction_label_column_name="predicted",
    actual_label_column_name="target",
    feature_column_names=[
        "fico_score",
        "merchant_id",
        "loan_amount",
        "annual_income",
        "home_ownership",
        "num_credit_lines",
        "inquests_in_last_6_months",
        "months_since_last_delinquency",
    ],
    tag_column_names=[
        "age",
        "gender",
    ],
)

Implicit Features

If your data has a large number of features, it can be inconvenient to list them all. For example, the breast cancer dataset below contains 30 features that can be used to predict whether a breast mass is malignant or benign. Instead of explicitly listing each feature, you can leave the feature_column_names field of your schema set to its default value of None, in which case, any columns of your dataframe that do not appear in your schema are implicitly assumed to be features.

Dataframe

target	predicted	mean radius	mean texture	mean perimeter	mean area	mean smoothness	mean compactness	mean concavity	mean concave points	mean symmetry	mean fractal dimension	radius error	texture error	perimeter error	area error	smoothness error	compactness error	concavity error	concave points error	symmetry error	fractal dimension error	worst radius	worst texture	worst perimeter	worst area	worst smoothness	worst compactness	worst concavity	worst concave points	worst symmetry	worst fractal dimension
malignant	benign	15.49	19.97	102.40	744.7	0.11600	0.15620	0.18910	0.09113	0.1929	0.06744	0.6470	1.3310	4.675	66.91	0.007269	0.02928	0.04972	0.01639	0.01852	0.004232	21.20	29.41	142.10	1359.0	0.1681	0.3913	0.55530	0.21210	0.3187	0.10190
malignant	malignant	17.01	20.26	109.70	904.3	0.08772	0.07304	0.06950	0.05390	0.2026	0.05223	0.5858	0.8554	4.106	68.46	0.005038	0.01503	0.01946	0.01123	0.02294	0.002581	19.80	25.05	130.00	1210.0	0.1111	0.1486	0.19320	0.10960	0.3275	0.06469
malignant	malignant	17.99	10.38	122.80	1001.0	0.11840	0.27760	0.30010	0.14710	0.2419	0.07871	1.0950	0.9053	8.589	153.40	0.006399	0.04904	0.05373	0.01587	0.03003	0.006193	25.38	17.33	184.60	2019.0	0.1622	0.6656	0.71190	0.26540	0.4601	0.11890
benign	benign	14.53	13.98	93.86	644.2	0.10990	0.09242	0.06895	0.06495	0.1650	0.06121	0.3060	0.7213	2.143	25.70	0.006133	0.01251	0.01615	0.01136	0.02207	0.003563	15.80	16.93	103.10	749.9	0.1347	0.1478	0.13730	0.10690	0.2606	0.07810
benign	benign	10.26	14.71	66.20	321.6	0.09882	0.09159	0.03581	0.02037	0.1633	0.07005	0.3380	2.5090	2.394	19.33	0.017360	0.04671	0.02611	0.01296	0.03675	0.006758	10.88	19.48	70.89	357.1	0.1360	0.1636	0.07162	0.04074	0.2434	0.08488

Schema

schema = px.Schema(
    prediction_label_column_name="predicted",
    actual_label_column_name="target",
)

Excluded Columns

You can tell Phoenix to ignore certain columns of your dataframe when implicitly inferring features by adding those column names to the excluded_column_names field of your schema. The dataframe below contains all the same data as the breast cancer dataset above, in addition to “hospital” and “insurance_provider” fields that are not features of your model. Explicitly exclude these fields, otherwise, Phoenix will assume that they are features.

Dataframe

target	predicted	hospital	insurance_provider	mean radius	mean texture	mean perimeter	mean area	mean smoothness	mean compactness	mean concavity	mean concave points	mean symmetry	mean fractal dimension	radius error	texture error	perimeter error	area error	smoothness error	compactness error	concavity error	concave points error	symmetry error	fractal dimension error	worst radius	worst texture	worst perimeter	worst area	worst smoothness	worst compactness	worst concavity	worst concave points	worst symmetry	worst fractal dimension
malignant	benign	Pacific Clinics	uninsured	15.49	19.97	102.40	744.7	0.11600	0.15620	0.18910	0.09113	0.1929	0.06744	0.6470	1.3310	4.675	66.91	0.007269	0.02928	0.04972	0.01639	0.01852	0.004232	21.20	29.41	142.10	1359.0	0.1681	0.3913	0.55530	0.21210	0.3187	0.10190
malignant	malignant	Queens Hospital	Anthem Blue Cross	17.01	20.26	109.70	904.3	0.08772	0.07304	0.06950	0.05390	0.2026	0.05223	0.5858	0.8554	4.106	68.46	0.005038	0.01503	0.01946	0.01123	0.02294	0.002581	19.80	25.05	130.00	1210.0	0.1111	0.1486	0.19320	0.10960	0.3275	0.06469
malignant	malignant	St. Francis Memorial Hospital	Blue Shield of CA	17.99	10.38	122.80	1001.0	0.11840	0.27760	0.30010	0.14710	0.2419	0.07871	1.0950	0.9053	8.589	153.40	0.006399	0.04904	0.05373	0.01587	0.03003	0.006193	25.38	17.33	184.60	2019.0	0.1622	0.6656	0.71190	0.26540	0.4601	0.11890
benign	benign	Pacific Clinics	Kaiser Permanente	14.53	13.98	93.86	644.2	0.10990	0.09242	0.06895	0.06495	0.1650	0.06121	0.3060	0.7213	2.143	25.70	0.006133	0.01251	0.01615	0.01136	0.02207	0.003563	15.80	16.93	103.10	749.9	0.1347	0.1478	0.13730	0.10690	0.2606	0.07810
benign	benign	CityMed	Anthem Blue Cross	10.26	14.71	66.20	321.6	0.09882	0.09159	0.03581	0.02037	0.1633	0.07005	0.3380	2.5090	2.394	19.33	0.017360	0.04671	0.02611	0.01296	0.03675	0.006758	10.88	19.48	70.89	357.1	0.1360	0.1636	0.07162	0.04074	0.2434	0.08488

Schema

schema = px.Schema(
    prediction_label_column_name="predicted",
    actual_label_column_name="target",
    excluded_column_names=[
        "hospital",
        "insurance_provider",
    ],
)

Embedding Features

Embedding features consist of vector data in addition to any unstructured data in the form of text or images that the vectors represent. Unlike normal features, a single embedding feature may span multiple columns of your dataframe. Use px.EmbeddingColumnNames to associate multiple dataframe columns with the same embedding feature.

• For a conceptual overview of embeddings, see Embeddings.

• For a comprehensive description of px.EmbeddingColumnNames, see the API reference.

The example in this section contain low-dimensional embeddings for the sake of easy viewing. Your embeddings in practice will typically have much higher dimension.

Embedding Vectors

To define an embedding feature, you must at minimum provide Phoenix with the embedding vector data itself. Specify the dataframe column that contains this data in the vector_column_name field on px.EmbeddingColumnNames. For example, the dataframe below contains tabular credit card transaction data in addition to embedding vectors that represent each row. Notice that:

• Unlike other fields that take strings or lists of strings, the argument to embedding_feature_column_names is a dictionary.

• The key of this dictionary, “transaction_embedding,” is not a column of your dataframe but is name you choose for your embedding feature that appears in the UI.

• The values of this dictionary are instances of px.EmbeddingColumnNames.

• Each entry in the “embedding_vector” column is a list of length 4.

Dataframe

predicted	target	embedding_vector	fico_score	merchant_id	loan_amount	annual_income	home_ownership	num_credit_lines	inquests_in_last_6_months	months_since_last_delinquency
fraud	not_fraud	[-0.97, 3.98, -0.03, 2.92]	604	Leannon Ward	22000	100781	RENT	108	0	0
fraud	not_fraud	[3.20, 3.95, 2.81, -0.09]	612	Scammeds	7500	116184	MORTGAGE	42	2	56
not_fraud	not_fraud	[-0.49, -0.62, 0.08, 2.03]	646	Leannon Ward	32000	73666	RENT	131	0	0
not_fraud	not_fraud	[1.69, 0.01, -0.76, 3.64]	560	Kirlin and Sons	19000	38589	MORTGAGE	131	0	0
uncertain	uncertain	[1.46, 0.69, 3.26, -0.17]	636	Champlin and Sons	10000	100251	MORTGAGE	10	0	3

Schema

schema = px.Schema(
    prediction_label_column_name="predicted",
    actual_label_column_name="target",
    embedding_feature_column_names={
        "transaction_embeddings": px.EmbeddingColumnNames(
            vector_column_name="embedding_vector"
        ),
    },
)

The features in this example are implicitly inferred to be the columns of the dataframe that do not appear in the schema.

To compare embeddings, Phoenix uses metrics such as Euclidean distance that can only be computed between vectors of the same length. Ensure that all embedding vectors for a particular embedding feature are one-dimensional arrays of the same length, otherwise, Phoenix will throw an error.

Embeddings of Images

If your embeddings represent images, you can provide links or local paths to image files you want to display in the app by using the link_to_data_column_name field on px.EmbeddingColumnNames. The following example contains data for an image classification model that detects product defects on an assembly line.

Dataframe

defective	image	image_vector
okay	https://www.example.com/image0.jpeg	[1.73, 2.67, 2.91, 1.79, 1.29]
defective	https://www.example.com/image1.jpeg	[2.18, -0.21, 0.87, 3.84, -0.97]
okay	https://www.example.com/image2.jpeg	[3.36, -0.62, 2.40, -0.94, 3.69]
defective	https://www.example.com/image3.jpeg	[2.77, 2.79, 3.36, 0.60, 3.10]
okay	https://www.example.com/image4.jpeg	[1.79, 2.06, 0.53, 3.58, 0.24]

Schema

schema = px.Schema(
    actual_label_column_name="defective",
    embedding_feature_column_names={
        "image_embedding": px.EmbeddingColumnNames(
            vector_column_name="image_vector",
            link_to_data_column_name="image",
        ),
    },
)

Local Images

For local image data, we recommend the following steps to serve your images via a local HTTP server:

1. In your terminal, navigate to a directory containing your image data and run python -m http.server 8000.

2. Add URLs of the form “http://localhost:8000/rel/path/to/image.jpeg” to the appropriate column of your dataframe.

For example, suppose your HTTP server is running in a directory with the following contents:

.
└── image-data
    └── example_image.jpeg

Then your image URL would be http://localhost:8000/image-data/example_image.jpeg.

Embeddings of Text

If your embeddings represent pieces of text, you can display that text in the app by using the raw_data_column_name field on px.EmbeddingColumnNames. The embeddings below were generated by a sentiment classification model trained on product reviews.

Dataframe

name	text	text_vector	category	sentiment
Magic Lamp	Makes a great desk lamp!	[2.66, 0.89, 1.17, 2.21]	office	positive
Ergo Desk Chair	This chair is pretty comfortable, but I wish it had better back support.	[3.33, 1.14, 2.57, 2.88]	office	neutral
Cloud Nine Mattress	I’ve been sleeping like a baby since I bought this thing.	[2.5, 3.74, 0.04, -0.94]	bedroom	positive
Dr. Fresh’s Spearmint Toothpaste	Avoid at all costs, it tastes like soap.	[1.78, -0.24, 1.37, 2.6]	personal_hygiene	negative
Ultra-Fuzzy Bath Mat	Cheap quality, began fraying at the edges after the first wash.	[2.71, 0.98, -0.22, 2.1]	bath	negative

Schema

schema = px.Schema(
    actual_label_column_name="sentiment",
    feature_column_names=[
        "category",
    ],
    tag_column_names=[
        "name",
    ],
    embedding_feature_column_names={
        "product_review_embeddings": px.EmbeddingColumnNames(
            vector_column_name="text_vector",
            raw_data_column_name="text",
        ),
    },
)

Multiple Embedding Features

Sometimes it is useful to have more than one embedding feature. The example below shows a multi-modal application in which one embedding represents the textual description and another embedding represents the image associated with products on an e-commerce site.

Dataframe

name	description	description_vector	image	image_vector
Magic Lamp	Enjoy the most comfortable setting every time for working, studying, relaxing or getting ready to sleep.	[2.47, -0.01, -0.22, 0.93]	https://www.example.com/image0.jpeg	[2.42, 1.95, 0.81, 2.60, 0.27]
Ergo Desk Chair	The perfect mesh chair, meticulously developed to deliver maximum comfort and high quality.	[-0.25, 0.07, 2.90, 1.57]	https://www.example.com/image1.jpeg	[3.17, 2.75, 1.39, 0.44, 3.30]
Cloud Nine Mattress	Our Cloud Nine Mattress combines cool comfort with maximum affordability.	[1.36, -0.88, -0.45, 0.84]	https://www.example.com/image2.jpeg	[-0.22, 0.87, 1.10, -0.78, 1.25]
Dr. Fresh’s Spearmint Toothpaste	Natural toothpaste helps remove surface stains for a brighter, whiter smile with anti-plaque formula	[-0.39, 1.29, 0.92, 2.51]	https://www.example.com/image3.jpeg	[1.95, 2.66, 3.97, 0.90, 2.86]
Ultra-Fuzzy Bath Mat	The bath mats are made up of 1.18-inch height premium thick, soft and fluffy microfiber, making it great for bathroom, vanity, and master bedroom.	[0.37, 3.22, 1.29, 0.65]	https://www.example.com/image4.jpeg	[0.77, 1.79, 0.52, 3.79, 0.47]

Schema

schema = px.Schema(
    tag_column_names=["name"],
    embedding_feature_column_names={
        "description_embedding": px.EmbeddingColumnNames(
            vector_column_name="description_vector",
            raw_data_column_name="description",
        ),
        "image_embedding": px.EmbeddingColumnNames(
            vector_column_name="image_vector",
            link_to_data_column_name="image",
        ),
    },
)

Distinct embedding features may have embedding vectors of differing length. The text embeddings in the above example have length 4 while the image embeddings have length 5.