do inference with ocean compute job

Author: devinaconley

README.md

@@ -29,5 +29,10 @@ Once the trained model has been uploaded to IPFS, define the `IPFS_MODEL_HASH` v
 
 Deploy model to web3 infrastructure
 ```
-python deploy --infra ocean
+python deploy.py --infra ocean
+```
+
+Make predictions with deployed model
+```
+python predict.py --infra ocean
 ```

deploy.py

@@ -1,5 +1,5 @@
 """
-deployed trained model to web3 infra
+deploy trained model to web3 infra
 """
 
 from mlweb3.ocean.deployment import deploy

mlweb3/ocean/inference.py

@@ -0,0 +1,106 @@
+"""
+inference logic for ocean model
+"""
+import os
+import time
+import datetime
+
+from dotenv import load_dotenv
+
+from brownie.network import accounts
+from web3 import Web3
+from ocean_lib.web3_internal.utils import connect_to_network
+from ocean_lib.example_config import ExampleConfig
+from ocean_lib.ocean.ocean import Ocean
+from ocean_lib.models.compute_input import ComputeInput
+
+
+def predict():
+    # configure client
+    load_dotenv()
+    connect_to_network('mumbai')
+    config = ExampleConfig.get_config('mumbai')
+    ocean = Ocean(config)
+    print(config)
+
+    # load accounts
+    accounts.clear()
+    alice_private_key = os.getenv('PRIVATE_KEY_0')
+    alice_wallet = accounts.add(alice_private_key)
+    alice_balance = accounts.at(alice_wallet.address).balance()
+    print('Alice balance: {}'.format(alice_balance))
+    assert alice_balance > 0
+
+    bob_private_key = os.getenv('PRIVATE_KEY_1')
+    bob_wallet = accounts.add(bob_private_key)
+    bob_balance = accounts.at(bob_wallet.address).balance()
+    print('Bob balance: {}'.format(bob_balance))
+    assert bob_balance > 0
+
+    # resolve assets
+    data_asset = ocean.assets.resolve('did:op:9a4c1fe5fcec7d4071b8afdbb17dcc2c05b1ecf8cf512f556ef8f38116a594c3')
+    weights_asset = ocean.assets.resolve('did:op:733ca2bfe560f6a66cc1e5cc2b6b91799db77e5fdbd41d4e1a402cb944bcdf43')
+    algo_asset = ocean.assets.resolve('did:op:0c3c2c9099c67a49128379e5781e48fccb6125d335464d07edcceae78e7f729c')
+
+    data_token = ocean.get_datatoken(data_asset.datatokens[0]['address'])
+    weights_token = ocean.get_datatoken(weights_asset.datatokens[0]['address'])
+    algo_token = ocean.get_datatoken(algo_asset.datatokens[0]['address'])
+
+    # mint tokens to user
+    data_token.mint(bob_wallet.address, Web3.toWei(5, 'ether'), {'from': alice_wallet})
+    weights_token.mint(bob_wallet.address, Web3.toWei(5, 'ether'), {'from': alice_wallet})
+    algo_token.mint(bob_wallet.address, Web3.toWei(5, 'ether'), {'from': alice_wallet})
+
+    # setup environment and inputs
+    data_service = data_asset.services[0]
+    weights_service = weights_asset.services[0]
+    algo_service = algo_asset.services[0]
+    free_c2d_env = ocean.compute.get_free_c2d_environment(data_service.service_endpoint)
+
+    data_compute_input = ComputeInput(data_asset, data_service)
+    weights_compute_input = ComputeInput(weights_asset, weights_service)
+    algo_compute_input = ComputeInput(algo_asset, algo_service)
+
+    # data_compute_input.transfer_tx_id = '0x8f9eb8f28402acf3d136edc063d2909059da2d418390eff52977f44548895197'
+    # weights_compute_input.transfer_tx_id = '0x62c9b4ac3d0686d4354847f2f0be8a67b96b8e97a6582cfc675e12626279fc0d'
+    # algo_compute_input.transfer_tx_id = '0x8b89f8dbcc0b2d9790c25cd403e264f9ccd876513f2926ae56ed1090fba26623'
+
+    # pay for dataset, weights, and algo for 1 day
+    datasets, algorithm = ocean.assets.pay_for_compute_service(
+        datasets=[data_compute_input, weights_compute_input],
+        algorithm_data=algo_compute_input,
+        consume_market_order_fee_address=bob_wallet.address,
+        wallet=bob_wallet,
+        compute_environment=free_c2d_env['id'],
+        valid_until=int((datetime.datetime.utcnow() + datetime.timedelta(days=1)).timestamp()),
+        consumer_address=free_c2d_env['consumerAddress'],
+    )
+    assert datasets, 'payment for dataset unsuccessful'
+    assert algorithm, 'payment for algorithm unsuccessful'
+
+    # start compute job
+    t0 = datetime.datetime.utcnow()
+    job_id = ocean.compute.start(
+        consumer_wallet=bob_wallet,
+        dataset=data_compute_input,
+        compute_environment=free_c2d_env['id'],
+        algorithm=algo_compute_input,
+        additional_datasets=[weights_compute_input]
+    )
+    print('Started compute job: {}'.format(job_id))
+
+    # monitor job
+    while 1:
+        status = ocean.compute.status(data_asset, data_service, job_id, bob_wallet)
+        t1 = datetime.datetime.utcnow()
+        print('status: {}, time elapsed: {} seconds'.format(status['statusText'], t1 - t0))
+        if status.get('dateFinished'):
+            print(status)
+            break
+        time.sleep(10)
+
+    # Retrieve algorithm output and log files
+    output = ocean.compute.compute_job_result_logs(
+        data_asset, data_service, job_id, bob_wallet
+    )
+    print(output)

predict.py

@@ -0,0 +1,13 @@
+"""
+make predictions with deployed model on web3 infra
+"""
+
+from mlweb3.ocean.inference import predict
+
+
+def main():
+    predict()
+
+
+if __name__ == '__main__':
+    main()