Revert "tensorboard dir tweaks"

This reverts commit 5393336.
awarebayes · Aug 7, 2020 · b76b42f · b76b42f
1 parent 51976d2
commit b76b42f
Show file tree

Hide file tree

Showing 17 changed files with 443 additions and 220 deletions.
diff --git a/examples/1. Vanilla RL/2. DDPG.ipynb b/examples/1. Vanilla RL/2. DDPG.ipynb
diff --git a/examples/1. Vanilla RL/3. TD3.ipynb b/examples/1. Vanilla RL/3. TD3.ipynb
diff --git a/examples/1. Vanilla RL/4. SAC.ipynb b/examples/1. Vanilla RL/4. SAC.ipynb
@@ -31,7 +31,6 @@
     "import numpy as np\n",
     "import pandas as pd\n",
     "from tqdm.auto import tqdm\n",
-    "from datetime import datetime\n",
     "\n",
     "import torch\n",
     "import torch.nn as nn\n",
@@ -420,7 +419,7 @@
     "    'test': {'value': [], 'softq': [], 'policy': [], 'step': []}\n",
     "    }\n",
     "\n",
-    "writer = SummaryWriter(log_dir='../../runs/SAC' + datetime.now().strftime(\"%H_%M\") + \"/\")\n",
+    "writer = SummaryWriter(log_dir='../../runs')\n",
     "debugger = recnn.Debugger(layout, run_tests, writer)\n",
     "plotter = recnn.Plotter(debugger, [['policy'], ['value', 'softq']],)\n",
     "torch.autograd.set_detect_anomaly(True)"

diff --git a/examples/2. REINFORCE TopK Off Policy Correction/1. Basic Reinforce with RecNN.ipynb b/examples/2. REINFORCE TopK Off Policy Correction/1. Basic Reinforce with RecNN.ipynb
diff --git a/examples/2. REINFORCE TopK Off Policy Correction/2. Reinforce Off Policy Correction.ipynb b/examples/2. REINFORCE TopK Off Policy Correction/2. Reinforce Off Policy Correction.ipynb
@@ -33,7 +33,6 @@
     "import matplotlib.pyplot as plt\n",
     "%matplotlib inline\n",
     "\n",
-    "from datetime import datetime\n",
     "\n",
     "# == recnn ==\n",
     "import sys\n",
@@ -195,7 +194,7 @@
     "reinforce = recnn.nn.Reinforce(policy_net, value_net)\n",
     "reinforce = reinforce.to(cuda)\n",
     "\n",
-    "reinforce.writer = SummaryWriter(log_dir='../../runs/ReinforceOffPolicy'+datetime.now().strftime(\"%H_%M\") + \"/\")\n",
+    "reinforce.writer = SummaryWriter(log_dir='../../runs')\n",
     "plotter = recnn.utils.Plotter(reinforce.loss_layout, [['value', 'policy']],)"
    ]
   },

diff --git a/...les/2. REINFORCE TopK Off Policy Correction/3. TopK Reinforce Off Policy Correction.ipynb b/...les/2. REINFORCE TopK Off Policy Correction/3. TopK Reinforce Off Policy Correction.ipynb
@@ -34,8 +34,6 @@
     "import matplotlib.pyplot as plt\n",
     "%matplotlib inline\n",
     "\n",
-    "from datetime import datetime\n",
-    "\n",
     "\n",
     "# == recnn ==\n",
     "import sys\n",
@@ -197,7 +195,7 @@
     "reinforce = recnn.nn.Reinforce(policy_net, value_net)\n",
     "reinforce = reinforce.to(cuda)\n",
     "\n",
-    "reinforce.writer = SummaryWriter(log_dir='../../runs/ReinforceTopKOffPolicy' + datetime.now().strftime(\"%H_%M\") + \"/\")\n",
+    "reinforce.writer = SummaryWriter(log_dir='../../runs')\n",
     "plotter = recnn.utils.Plotter(reinforce.loss_layout, [['value', 'policy']],)"
    ]
   },

diff --git a/examples/[Library Basics]/algorithms how to/ddpg.ipynb b/examples/[Library Basics]/algorithms how to/ddpg.ipynb
diff --git a/examples/[Results]/1. Ranking.ipynb → examples/_ Results/1. Ranking.ipynb b/examples/[Results]/1. Ranking.ipynb → examples/_ Results/1. Ranking.ipynb
diff --git a/...esults]/2. Diversity Test (Indexes).ipynb → ...Results/2. Diversity Test (Indexes).ipynb b/...esults]/2. Diversity Test (Indexes).ipynb → ...Results/2. Diversity Test (Indexes).ipynb
diff --git a/examples/[Results]/3. Distances Test.ipynb → examples/_ Results/3. Distances Test.ipynb b/examples/[Results]/3. Distances Test.ipynb → examples/_ Results/3. Distances Test.ipynb
diff --git a/...sults]/4. BCQ Stochastic Diversity .ipynb → ...esults/4. BCQ Stochastic Diversity .ipynb b/...sults]/4. BCQ Stochastic Diversity .ipynb → ...esults/4. BCQ Stochastic Diversity .ipynb
diff --git a/readme.md b/readme.md
@@ -42,14 +42,15 @@ This is my school project. It focuses on Reinforcement Learning for personalized
 
 - The learning is built around sequential or frame environment that supports ML20M and like. Seq and Frame determine the length type of sequential data, seq is fully sequential dynamic size, while the frame is just a static frame.
 
-- State Representation module with various methods. For sequential state representation, you can use LSTM/RNN/GRU (WIP)
+- State Representation module with various methods. For sequential state representation, you can use basic LSTM/RNN/GRU, 
+Temporal Convolutional Networks, Echo State Networks and Chaos Free RNNs that are way faster than GRU.
 
-- Parallel data loading with Modin (Dask and Ray) with one line of code = 3x time improvement
-
-- Pytorch 1.5 support with Tensorboard visualization.
+- Pytorch 1.3 support with Tensorboard visualization.
 
 - New datasets will be added in the future.
 
+- SOTA optimizers (RAdam, LookAhead, Ranger) come pre-packaged.
+
 ## 📚 Medium Articles
 
 The repo consists of two parts: the library (./recnn), and the playground (./examples) where I explain how to work with certain things. 
@@ -88,8 +89,10 @@ The repo consists of two parts: the library (./recnn), and the playground (./exa
 
 ### ‍Repos I used code from:
 
-- Sfujim's [BCQ](https://github.com/sfujim/BCQ) (not released yet)
-- Higgsfield's [RL Adventure 2](https://github.com/higgsfield/RL-Adventure-2) (great insiration)
+- Sfujim's [BCQ](https://github.com/sfujim/BCQ)
+- LiyuanLucasLiu [Radam](https://github.com/LiyuanLucasLiu/RAdam)
+- Higgsfield's [RL Adventure 2](https://github.com/higgsfield/RL-Adventure-2)
+- Mpyrozhok's [adamwr](https://github.com/mpyrozhok/adamwr)
 
 ### 🤔 What is this? 
 
@@ -136,7 +139,7 @@ recommendations, pairwise distances, and pinpoint accuracy.
 
 Run it:
 ```
-git clone --depth 1 git@github.com:awarebayes/RecNN.git 
+git clone git@github.com:awarebayes/RecNN.git 
 cd RecNN && streamlit run examples/streamlit_demo.py
 ```
 

diff --git a/recnn/__init__.py b/recnn/__init__.py
@@ -1,2 +1 @@
 from recnn import data, utils, nn, rep
-from recnn.data import pd as pd
diff --git a/recnn/data/__init__.py b/recnn/data/__init__.py
@@ -2,4 +2,3 @@
 from .utils import *
 from .env import *
 from .dataset_functions import *
-from .pandas_backend import pd
diff --git a/recnn/data/dataset_functions.py b/recnn/data/dataset_functions.py
@@ -1,5 +1,4 @@
 from recnn.data.utils import make_items_tensor
-from .pandas_backend import pd
 
 """     
     What?
@@ -42,12 +41,6 @@ def prepare_dataset(**kwargs):
     build_data_pipeline function, and it is passed down the function chain. If needed, it will be used. Otherwise, ignored  
 """
 
-def try_progress_apply(data_frame, function):
-    try: 
-        return data_frame.progress_apply(function)
-    except AttributeError:
-        return data_frame.apply(function)
-
 
 def prepare_dataset(df, key_to_id, frame_size, env, sort_users=False, **kwargs):
 
@@ -56,28 +49,26 @@ def prepare_dataset(df, key_to_id, frame_size, env, sort_users=False, **kwargs):
         [1, 34, 123, 2000], recnn makes it look like [0,1,2,3] for you.
     """
 
-    df['rating'] =  try_progress_apply(df['rating'], lambda i: 2 * (i - 2.5))
-    df['movieId'] = try_progress_apply(df['movieId'], lambda i: key_to_id.get(i))
+    df['rating'] = df['rating'].progress_apply(lambda i: 2 * (i - 2.5))
+    df['movieId'] = df['movieId'].progress_apply(lambda i: key_to_id.get(i))
 
     users = df[['userId', 'movieId']].groupby(['userId']).size()
     users = users[users > frame_size]
     if sort_users:
         users = users.sort_values(ascending=False)
     users = users.index
-    if pd.get_type() == "modin": # modin groupby.apply isn't much faster
-        df = df._to_pandas() # and collect better be single threaded
     ratings = df.sort_values(by='timestamp').set_index('userId').drop('timestamp', axis=1).groupby('userId')
-    
+
     # Groupby user
     user_dict = {}
 
     def app(x):
-        userid = int(x.index[0])
-        user_dict[userid] = {}
-        user_dict[userid]['items'] = x['movieId'].values
-        user_dict[userid]['ratings'] = x['rating'].values
+        userid = x.index[0]
+        user_dict[int(userid)] = {}
+        user_dict[int(userid)]['items'] = x['movieId'].values
+        user_dict[int(userid)]['ratings'] = x['rating'].values
 
-    try_progress_apply(ratings, app)
+    ratings.progress_apply(app)
 
     env.user_dict = user_dict
     env.users = users

diff --git a/recnn/data/env.py b/recnn/data/env.py
@@ -1,6 +1,6 @@
 from . import utils, dataset_functions as dset_F
 import pickle
-from .pandas_backend import pd
+import pandas as pd
 from torch.utils.data import Dataset, DataLoader
 import torch
 import numpy as np
@@ -101,8 +101,7 @@ def __init__(self, embeddings, ratings, test_size=0.05, min_seq_size=10,
         self.embeddings = movies_embeddings_tensor
         self.key_to_id = key_to_id
         self.id_to_key = id_to_key
-        self.ratings = pd.get().read_csv(ratings)
-
+        self.ratings = pd.read_csv(ratings)
 
         self.user_dict = None
         self.users = None  # filtered keys of user_dict

diff --git a/recnn/data/utils.py b/recnn/data/utils.py
@@ -1,7 +1,7 @@
 
 import numpy as np
 import torch
-from .pandas_backend import pd
+import pandas as pd
 
 # helper function similar to pandas.Series.rolling
 def rolling_window(a, window):
@@ -112,7 +112,7 @@ def padder(x):
 
 
 def sort_users_itemwise(user_dict, users):
-    return pd.get().Series(dict([(i, user_dict[i]['items'].shape[0]) for i in users])).sort_values(ascending=False).index
+    return pd.Series(dict([(i, user_dict[i]['items'].shape[0]) for i in users])).sort_values(ascending=False).index
 
 
 def prepare_batch_dynamic_size(batch, item_embeddings_tensor, embed_batch=None):