Instructions to run locally; close #2, close #4

Author: mantasu

.gitignore

@@ -12,6 +12,7 @@ assignment1/assignment.pdf
 
 # Assignment 2 generated
 assignment2/cs231n/build
+assignment2/cs231n/im2col_cython.cpython*
 assignment2/cs231n/datasets/cifar-10-batches-py
 assignment2/a2.zip
 assignment2/assignment.pdf

README.md

@@ -1,12 +1,12 @@
 <h1 align="center">CS231n: Assignment Solutions</h1>
 <p align="center"><b>Convolutional Neural Networks for Visual Recognition</b></p>
-<p align="center"><i>Stanford - Spring 2021</i></p>
+<p align="center"><i>Stanford - Spring 2021-2023</i></p>
 
 ## About
 ### Overview
-These are my solutions for the **CS231n** course assignments offered by Stanford University (Spring 2021). Solutions work for further years like 2022. Inline questions are explained in detail, the code is brief and commented (see examples below). From what I investigated, these should be the shortest code solutions (excluding open-ended challenges). In assignment 2, _DenseNet_ is used in _PyTorch_ notebook and _ResNet_ in _TensorFlow_ notebook. 
+These are my solutions for the **CS231n** course assignments offered by Stanford University (Spring 2021). Solutions work for further years like 2022, 2023. Inline questions are explained in detail, the code is brief and commented (see examples below). From what I investigated, these should be the shortest code solutions (excluding open-ended challenges). In assignment 2, _DenseNet_ is used in _PyTorch_ notebook and _ResNet_ in _TensorFlow_ notebook. 
 
-> Check out the solutions for **[CS224n](https://github.com/mantasu/cs224n)**. From what I checked, they contain more comprehensive explanations than others.
+> Check out the solutions for **[CS224n](https://github.com/mantasu/cs224n)**. They contain more comprehensive explanations than others.
 
 ### Main sources (official)
 * [**Course page**](http://cs231n.stanford.edu/index.html)
@@ -42,7 +42,33 @@ These are my solutions for the **CS231n** course assignments offered by Stanford
 
 <br>
 
+## Running Locally
+
+It is advised to run in [Colab](https://colab.research.google.com/), however, you can also run locally. To do so, first, set up your environment - either through [conda](https://docs.conda.io/en/latest/) or [venv](https://docs.python.org/3/library/venv.html). It is advised to install [PyTorch](https://pytorch.org/get-started/locally/) in advance with GPU acceleration. Then, follow the steps:
+1. Install the required packages:
+   ```bash
+   pip install -r requirements.txt
+   ```
+2. Change every first code cell in `.ipynb` files to:
+   ```bash
+   %cd cs231n/datasets/
+   !bash get_datasets.sh
+   %cd ../../
+   ```
+3. Change the first code cell in section **Fast Layers** in [ConvolutionalNetworks.ipynb](assignment2/ConvolutionalNetworks.ipynb) to:
+   ```bash
+   %cd cs231n
+   !python setup.py build_ext --inplace
+   %cd ..
+   ```
+
+I've gathered all the requirements for all 3 assignments into one file [requirements.txt](requirements.txt) so there is no need to additionally install the requirements specified under each assignment folder. If you plan to complete [TensorFlow.ipynb](assignment2/TensorFlow.ipynb), then you also need to additionally install [Tensorflow](https://www.tensorflow.org/install).
+
+
+> **Note**: to use MPS acceleration via Apple M1, see the comment in [#4](https://github.com/mantasu/cs231n/issues/4#issuecomment-1492202538).
+
 ## Examples
+
 <details><summary><b>Inline question example</b></summary>
 <br>
 <b>Inline Question 1</b>

assignment1/cs231n/classifiers/k_nearest_neighbor.py

@@ -1,7 +1,6 @@
 from builtins import range
 from builtins import object
 import numpy as np
-from past.builtins import xrange
 
 
 class KNearestNeighbor(object):

assignment1/cs231n/classifiers/linear_classifier.py

@@ -5,7 +5,6 @@
 import numpy as np
 from ..classifiers.linear_svm import *
 from ..classifiers.softmax import *
-from past.builtins import xrange
 
 
 class LinearClassifier(object):

assignment1/cs231n/classifiers/linear_svm.py

@@ -1,7 +1,6 @@
 from builtins import range
 import numpy as np
 from random import shuffle
-from past.builtins import xrange
 
 
 def svm_loss_naive(W, X, y, reg):

assignment1/cs231n/classifiers/softmax.py

@@ -1,7 +1,6 @@
 from builtins import range
 import numpy as np
 from random import shuffle
-from past.builtins import xrange
 
 
 def softmax_loss_naive(W, X, y, reg):

assignment1/cs231n/features.py

@@ -1,7 +1,6 @@
 from __future__ import print_function
 from builtins import zip
 from builtins import range
-from past.builtins import xrange
 
 import matplotlib
 import numpy as np

assignment1/cs231n/gradient_check.py

@@ -1,6 +1,5 @@
 from __future__ import print_function
 from builtins import range
-from past.builtins import xrange
 
 import numpy as np
 from random import randrange

assignment1/cs231n/layers.py

@@ -149,8 +149,8 @@ class for the ith input.
     ###########################################################################
     # *****START OF YOUR CODE (DO NOT DELETE/MODIFY THIS LINE)*****
 
-    N = len(y)      # number of samples
-    x_true = x[range(N), y][:, None]    # scores for true labels
+    N = len(y)                                # number of samples
+    x_true = x[range(N), y][:, None]          # scores for true labels
     margins = np.maximum(0, x - x_true + 1)   # margin for each score
     loss = margins.sum() / N - 1
     dx = (margins > 0).astype(float) / N
@@ -187,8 +187,8 @@ class for the ith input.
 
     N = len(y) # number of samples
 
-    P = np.exp(x - x.max())              # numerically stable exponents
-    P /= P.sum(axis=1, keepdims=True)    # row-wise probabilities (softmax)
+    P = np.exp(x - x.max(axis=1, keepdims=True)) # numerically stable exponents
+    P /= P.sum(axis=1, keepdims=True)            # row-wise probabilities (softmax)
 
     loss = -np.log(P[range(N), y]).sum() / N # sum cross entropies as loss
 

assignment1/cs231n/solver.py

@@ -1,7 +1,5 @@
 from __future__ import print_function, division
-from future import standard_library
 
-standard_library.install_aliases()
 from builtins import range
 from builtins import object
 import os

assignment1/cs231n/vis_utils.py

@@ -1,5 +1,4 @@
 from builtins import range
-from past.builtins import xrange
 
 from math import sqrt, ceil
 import numpy as np

assignment1/knn.ipynb

@@ -1,6 +1,5 @@
 from __future__ import print_function
 from builtins import range
-from past.builtins import xrange
 
 import numpy as np
 from random import randrange

assignment2/cs231n/gradient_check.py

@@ -394,11 +394,10 @@ def layernorm_forward(x, gamma, beta, ln_param):
     ###########################################################################
     # *****START OF YOUR CODE (DO NOT DELETE/MODIFY THIS LINE)*****
 
-    ln_param.setdefault('mode', 'train')       # same as batchnorm in train mode
-    ln_param.setdefault('axis', 1)             # over which axis to sum for grad
-    [gamma, beta] = np.atleast_2d(gamma, beta) # assure 2D to perform transpose
+    bn_param = {"mode": "train", "axis": 1, **ln_param} # same as batchnorm in train mode + over which axis to sum for grad
+    [gamma, beta] = np.atleast_2d(gamma, beta)          # assure 2D to perform transpose
 
-    out, cache = batchnorm_forward(x.T, gamma.T, beta.T, ln_param) # same as batchnorm
+    out, cache = batchnorm_forward(x.T, gamma.T, beta.T, bn_param) # same as batchnorm
     out = out.T                                                    # transpose back
 
     # *****END OF YOUR CODE (DO NOT DELETE/MODIFY THIS LINE)*****
@@ -836,14 +835,14 @@ def spatial_groupnorm_forward(x, gamma, beta, G, gn_param):
     ###########################################################################
     # *****START OF YOUR CODE (DO NOT DELETE/MODIFY THIS LINE)*****
 
-    N, C, H, W = x.shape                                      # input dims
-    gn_param.update({'shape':(W, H, C, N), 'axis':(0, 1, 3)}) # params to reuse batchnorm method
+    N, C, H, W = x.shape                                            # input dims
+    ln_param = {"shape":(W, H, C, N), "axis":(0, 1, 3), **gn_param} # params to reuse batchnorm method
 
     x = x.reshape(N*G, -1)                                    # reshape x to use vanilla layernorm
     gamma = np.tile(gamma, (N, 1, H, W)).reshape(N*G, -1)     # reshape gamma to use vanilla layernorm
     beta = np.tile(beta, (N, 1, H, W)).reshape(N*G, -1)       # reshape beta to use vanilla layernorm
 
-    out, cache = layernorm_forward(x, gamma, beta, gn_param)  # perform vanilla layernorm
+    out, cache = layernorm_forward(x, gamma, beta, ln_param)  # perform vanilla layernorm
     out = out.reshape(N, C, H, W)                             # reshape back the output
     cache = (G, cache)                                        # cache involves G
 

assignment2/cs231n/layers.py

@@ -1,7 +1,5 @@
 from __future__ import print_function, division
-from future import standard_library
 
-standard_library.install_aliases()
 from builtins import range
 from builtins import object
 import os

assignment2/cs231n/solver.py

@@ -1,5 +1,4 @@
 from builtins import range
-from past.builtins import xrange
 
 from math import sqrt, ceil
 import numpy as np

assignment2/cs231n/vis_utils.py

@@ -0,0 +1,11 @@
+Cython
+tqdm
+h5py
+torch
+torchvision
+tabulate
+scipy
+pandas
+imageio
+matplotlib
+ipykernel