fjcasti1

# CREATE NULL VALUES IN SOME ROWS FOR 1 OR MORE COLUMNS
import random
N_rows = 10 # How many rows?
index = random.sample(range(len(df)), N_rows) # Get random indexes for how many rows you specified
col_list = ['col-1','col-2',...] # Select columns to add Nones to
df.loc[index, col_list]=None

fjcasti1

# CREATE 1 OR MORE COLUMNS FULL OF NONES
col_list = ['col-1','col-2',...]
df[col_list]=None

fjcasti1

env_names = ['training', 'validation', 'production']
environments = [Environments.TRAINING, Environments.VALIDATION, Environments.PRODUCTION]
dfs = [train_df, val_df, prod_df]

# Logging DataFrames
for env_name, env, df in zip(env_names, environments, dfs):
    response = arize_client.log(
        dataframe=df,
        model_id=model_id,
        model_version=model_version,

fjcasti1

features = []
arize_columns = [
    'prediction_id',
    'prediction_ts',
    'label',
    'pred_label',
    'image_vector',
    'url'
]

fjcasti1

SPACE_KEY = "SPACE_KEY"
API_KEY = "API_KEY"
arize_client = Client(space_key=SPACE_KEY, api_key=API_KEY)
model_id = "CV-demo-fashion-mnist-quality-drift"
model_version = "1.0"
model_type = ModelTypes.SCORE_CATEGORICAL
if SPACE_KEY == "SPACE_KEY" or API_KEY == "API_KEY":
    raise ValueError("❌ NEED TO CHANGE SPACE AND/OR API_KEY")
else:
    print("✅ Import and Setup Arize Client Done! Now we can start using Arize!")

fjcasti1

train_df['label'] = train_df['label'].map(lambda label: id2label[label])
train_df['pred_label'] = train_df['pred_label'].map(lambda label: id2label[label])

val_df['label'] = val_df['label'].map(lambda label: id2label[label])
val_df['pred_label'] = val_df['pred_label'].map(lambda label: id2label[label])

prod_df['label'] = prod_df['label'].map(lambda label: id2label[label])
prod_df['pred_label'] = prod_df['pred_label'].map(lambda label: id2label[label])

fjcasti1

def postprocess(batch):
    inputs = feature_extractor([x.convert("RGB") for x in batch["image"]], return_tensors="pt").to(device)
    with torch.no_grad():
        outputs = model(**inputs)

    pred_labels = torch.argmax(outputs.logits, dim=1).cpu().numpy()

    last_hidden_states = outputs.hidden_states[-1]
    embeddings = torch.mean(last_hidden_states, (2, 3)).cpu().numpy()

fjcasti1

trainer = Trainer(
    model=model,
    args=training_args,
    data_collator=collate_fn,
    train_dataset=train_ds,
    eval_dataset=val_ds,
    tokenizer=feature_extractor,
    compute_metrics=compute_metrics,
)

fjcasti1

def collate_fn(dataset):
    pixel_values = torch.stack([ds["pixel_values"] for ds in dataset])
    labels = torch.tensor([ds["label"] for ds in dataset])
    return {"pixel_values": pixel_values, "labels": labels}

fjcasti1

def compute_metrics(pred):
    labels = pred.label_ids
    preds = pred.predictions[0].argmax(-1)
    f1 = f1_score(labels, preds, average="weighted")
    acc = accuracy_score(labels, preds)
    return {"accuracy": acc, "f1": f1}