add related notebook (#39)

notebooks/util/fcst_utils.py

@@ -0,0 +1,110 @@
+import time
+import boto3
+import json
+import pandas as pd
+import logging
+import matplotlib.pyplot as plt
+
+def wait_till_delete(callback, check_time = 5, timeout = 180):
+    elapsed_time = 0
+    while elapsed_time < timeout:
+        try:
+            out = callback()
+        except Exception as e:
+            # When given the resource not found exception, deletion has occured
+            if e.response['Error']['Code'] == 'ResourceNotFoundException':
+                logging.info('Successful delete\n')
+                return
+            # Fails with other error
+            logging.info(f'Deletion failed: {e}')
+            return(e)
+        time.sleep(check_time)  # units of seconds
+        elapsed_time += check_time
+
+def wait(callback, time_interval=30):
+    last_status = callback()['Status']
+    time.sleep(time_interval)
+    elapsed_time = time_interval
+    is_failed = True
+
+    while (last_status != 'ACTIVE'):
+        last_status = callback()['Status']
+        time.sleep(time_interval)  # units of seconds
+        elapsed_time += time_interval
+        print('.', end='', flush=True)
+        if last_status == 'CREATE_FAILED':
+            break
+    if last_status == "ACTIVE":
+        is_failed = False
+    job_status = "failed" if is_failed else "success"
+    print('')
+    logging.info(f"Finished in {elapsed_time} seconds with status {job_status}")
+    return not is_failed
+
+def load_exact_sol(fname, item_id):
+    exact = pd.read_csv(fname, header = None)
+    exact.columns = ['item_id', 'timestamp', 'target']
+    return exact.loc[exact['item_id'] == item_id]
+
+def get_or_create_role_arn():
+    iam = boto3.client("iam")
+    role_name = "ForecastRoleDemo"
+    assume_role_policy_document = {
+        "Version": "2012-10-17",
+        "Statement": [
+            {
+              "Effect": "Allow",
+              "Principal": {
+                "Service": "forecast.amazonaws.com"
+              },
+              "Action": "sts:AssumeRole"
+            }
+        ]
+    }
+    role_arn = None
+    try:
+        create_role_response = iam.create_role(
+            RoleName = role_name,
+            AssumeRolePolicyDocument = json.dumps(assume_role_policy_document)
+        )
+        role_arn = create_role_response["Role"]["Arn"]
+    except iam.exceptions.EntityAlreadyExistsException:
+        print("The role " + role_name + "exists, ignore to create it")
+        role_arn = boto3.resource('iam').Role(role_name).arn
+    # AmazonPersonalizeFullAccess provides access to any S3 bucket with a name that includes "personalize" or "Personalize"
+    # if you would like to use a bucket with a different name, please consider creating and attaching a new policy
+    # that provides read access to your bucket or attaching the AmazonS3ReadOnlyAccess policy to the role
+    policy_arn = "arn:aws:iam::aws:policy/AmazonForecastFullAccess"
+    iam.attach_role_policy(
+        RoleName = role_name,
+        PolicyArn = policy_arn
+    )
+
+    # Now add S3 support
+
+    iam.attach_role_policy(
+        PolicyArn='arn:aws:iam::aws:policy/AmazonS3FullAccess',
+        RoleName=role_name
+    )
+    time.sleep(60) # wait for a minute to allow IAM role policy attachment to propagate
+    print(role_arn)
+    return role_arn
+
+
+def plot_forecasts(fcsts, exact, freq = '1H', forecastHorizon=24, time_back = 80):
+    p10 = pd.DataFrame(fcsts['Forecast']['Predictions']['p10'])
+    p50 = pd.DataFrame(fcsts['Forecast']['Predictions']['p50'])
+    p90 = pd.DataFrame(fcsts['Forecast']['Predictions']['p90'])
+    pred_int = p50['Timestamp'].apply(lambda x: pd.Timestamp(x))
+    fcst_start_date = pred_int[0]
+    time_int = exact['timestamp'].apply(lambda x: pd.Timestamp(x))
+    plt.plot(time_int[-time_back:],exact['target'].values[-time_back:], color = 'r')
+    plt.plot(pred_int, p50['Value'].values, color = 'k');
+    plt.fill_between(p50['Timestamp'].values, 
+                     p10['Value'].values,
+                     p90['Value'].values,
+                     color='b', alpha=0.3);
+    plt.axvline(x=pd.Timestamp(fcst_start_date), linewidth=3, color='g', ls='dashed');
+    plt.axvline(x=pd.Timestamp(fcst_start_date, freq)+forecastHorizon-1, linewidth=3, color='g', ls='dashed');
+    plt.xticks(rotation=30);
+    plt.legend(['Target', 'Forecast'], loc = 'lower left')